| .. |
|
_shared
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
diffusion
|
Fix Nemotron loss normalization across gradient accumulation (#14002)
|
2026-10-18 08:18:10 -07:00 |
|
fast_inference
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
kaggle
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
notebooks
|
Notebooks smoke: run the install cell the way Colab does (#14058)
|
2026-10-17 01:40:50 -07:00 |
|
python
|
Studio: install from a package index that lags PyPI (#14133)
|
2026-10-18 09:50:33 -07:00 |
|
qlora
|
|
|
|
saving
|
Create the branch before pushing a LoRA GGUF to a revision (#13915)
|
2026-10-18 05:43:11 -07:00 |
|
security
|
CI: pin the Arrow and sqlite-vec tags the Windows ARM wheelhouse signs to commits (#14092)
|
2026-10-18 02:47:50 -07:00 |
|
sh
|
Studio: install from a package index that lags PyPI (#14133)
|
2026-10-18 09:50:33 -07:00 |
|
studio
|
Merge #10280: main merged in (unsloth doctor --mcp)
|
2026-10-18 18:39:38 -07:00 |
|
studio_setup_ps1
|
Desktop: explain Windows Long Paths before its UAC prompt (#13523)
|
2026-10-15 08:09:50 -07:00 |
|
utils
|
Fix packed document boundaries in Qwen PLE (#14011)
|
2026-10-18 09:26:08 -07:00 |
|
version_compat
|
Sync full finetune weights into colocated vLLM in a few load_weights calls (#13901)
|
2026-10-16 02:04:00 -07:00 |
|
vllm_compat
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
__init__.py
|
|
|
|
_grpo_dispatch_source.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
_nvfp4_fixtures.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
_rl_source.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
_torchcodec_stub.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
_zoo_aggressive_cuda_spoof.py
|
CUDA spoof: cover CUDA wheels on GPU-less runners, not only CPU wheels (#14056)
|
2026-10-17 01:23:58 -07:00 |
|
_zoo_rocm_spoof.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
conftest.py
|
Studio: default to the Vulkan llama.cpp bundle when a CPU-bound host has a Vulkan GPU (#13599)
|
2026-10-18 07:36:02 -07:00 |
|
run_all.sh
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_accelerator_distributed_type_patch.py
|
Train with DeepSpeed ZeRO-2 and ZeRO-3 (#13587)
|
2026-10-14 04:12:07 -07:00 |
|
test_accepts_loss_kwargs_head_walk.py
|
Parse each forward once in the per-step loss-kwargs walk (#13585)
|
2026-10-13 04:30:13 -07:00 |
|
test_accepts_loss_kwargs_remote_mean_loss.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_adapter_vocab_growth.py
|
Reload LoRA adapters trained with added tokens (#4219)
|
2026-09-29 15:33:26 -07:00 |
|
test_allow_cpu_import_driverless.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_amp_helpers_every_device.py
|
Let unsloth.models and unsloth.save import on Apple Silicon (#9447)
|
2026-09-07 23:22:21 -03:00 |
|
test_apertus_instruct_mapping.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_apply_qkv_fallback.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_async_grpo_fast_inference.py
|
Tests: keep five tests off what the host has installed or downloaded (#13395)
|
2026-10-12 03:01:37 -07:00 |
|
test_async_grpo_packed_rows.py
|
AsyncGRPOTrainer with fast_inference: train and generate on one GPU with shared weights (#13292)
|
2026-10-11 20:11:01 -07:00 |
|
test_async_grpo_patch.py
|
AsyncGRPOTrainer: optional SAO-style importance sampling mask (#13820)
|
2026-10-15 05:35:13 -07:00 |
|
test_attention_implementation.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_attn_dtype_custom_datatype.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_attn_impl_baked_nested_config.py
|
Carry the resolved attention implementation to nested configs a remote config baked flash attention into (Nemotron 3 Nano Omni) (#12099)
|
2026-09-27 04:01:42 -07:00 |
|
test_attn_impl_honor_explicit.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_bad_mappings_redirect.py
|
Load six older models in 4bit instead of a missing repo (#13941)
|
2026-10-18 06:50:48 -07:00 |
|
test_bailing_hybrid_low_rank_query.py
|
Load Ling 3.x low-rank queries on native bailing_hybrid and fix its exports (#14136)
|
2026-10-18 09:25:08 -07:00 |
|
test_beam_search_reorder_cache.py
|
Fix beam search on the fast decode path (#13053)
|
2026-10-09 04:52:19 -07:00 |
|
test_bf16_quant_disabled_message.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_block_compile.py
|
Keep opted-in bnb 4-bit MoE experts outside the compiled block (#13946)
|
2026-10-16 06:50:04 -07:00 |
|
test_block_swap_auto_ddp.py
|
Refuse an explicit offload_layers swap under FSDP / DeepSpeed (#13857)
|
2026-10-15 20:27:05 -07:00 |
|
test_block_swap_install.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_bnb_checkpoint_skip_modules.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_bnb_cpu_spill_message.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_bnb_integration_compile.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_bnb_nf4_override.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_bnb_optimizer_step_sync.py
|
Skip bitsandbytes' per-parameter device sync in non-paged optimizer steps (#13389)
|
2026-10-13 01:57:13 -07:00 |
|
test_bnb_paged_optimizer_resume.py
|
Keep paged bitsandbytes optimizer state paged on checkpoint resume (#13092)
|
2026-10-09 04:39:52 -07:00 |
|
test_bnb_prequantized_save.py
|
Save models loaded from pre-quantized bitsandbytes checkpoints on transformers 5 (#13054)
|
2026-10-08 08:56:35 -07:00 |
|
test_bnb_quantized_nothing_warning.py
|
Point -bf16 4-bit requests at quantization_config, and warn when a bitsandbytes load quantized nothing (#11586)
|
2026-09-25 11:05:44 -07:00 |
|
test_bnb_triton_quantize_fallback.py
|
Intel XPU: 4bit loads without a C compiler by falling back to bitsandbytes' PyTorch quantize (#13309)
|
2026-10-11 08:39:01 -07:00 |
|
test_broken_tf_does_not_break_import.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_broken_torchvision_probe.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_callback_signature_drift.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_chat_template_trailing_generation_prompt.py
|
Repair chat templates that always append the generation prompt, also on FastModel loads (#12139)
|
2026-09-28 02:06:43 -07:00 |
|
test_chat_template_whitespace.py
|
Stop the Starling, Yi-chat and LFM2 templates leaking whitespace (#11779)
|
2026-09-25 05:22:23 -07:00 |
|
test_chunked_mask_block_sequence_ids.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_ci_backend_pytest_shards.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_ci_repo_cpu_shards.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_cli_chat_compare_precision.py
|
Studio: load a 16-bit Continued Pretraining run in 16-bit in Chat (#11480)
|
2026-09-22 07:22:35 -07:00 |
|
test_cli_export_unpacking.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_cold_import_order.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_collection_hygiene.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_compile_decode_selection.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_compiled_auto_model.py
|
Load the compiled class when auto_model is a concrete class (#12757)
|
2026-10-05 17:27:56 -07:00 |
|
test_composite_prefix_renaming.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_compressed_export_gpu_release.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_compressed_tensors_activation_ste.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_compressed_tensors_bnb.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_compressed_tensors_fp8_kernel_route.py
|
FP8 GEMV on Blackwell + Triton 3.3: no graph break under torch 2.7, keep the exact GEMV shapes that are faster (#14026)
|
2026-10-16 10:52:12 -07:00 |
|
test_compressed_tensors_int4.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_compressed_tensors_load_decompress.py
|
Pass gradients through compressed-tensors activation quantization so W8A8 checkpoints train with LoRA (#11585)
|
2026-09-27 09:12:46 -07:00 |
|
test_construct_chat_template_processor.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_context_parallel.py
|
Context parallel: compose with FSDP2 on a (dp_shard, cp) mesh (#13910)
|
2026-10-16 01:33:28 -07:00 |
|
test_context_parallel_fsdp2_gloo.py
|
Context parallel: compose with FSDP2 on a (dp_shard, cp) mesh (#13910)
|
2026-10-16 01:33:28 -07:00 |
|
test_context_parallel_fsdp2_save_gloo.py
|
Context parallel: gather the FSDP2 state dict in save_model on transformers 4.x (#14028)
|
2026-10-16 11:21:09 -07:00 |
|
test_cross_entropy_all_masked_labels.py
|
fast_cross_entropy_loss: all-masked labels give loss 0 instead of NaN (#13325)
|
2026-10-13 11:53:21 -07:00 |
|
test_cross_entropy_eviction_hints.py
|
Kernel follow-ups: NVFP4 shape-aware fused gate, MoE fallback tiles, test fixes across torch 2.10 - 2.14 (#14122)
|
2026-10-18 08:22:25 -07:00 |
|
test_cross_entropy_input_layout.py
|
fix: handle strided cross entropy inputs (#10713)
|
2026-09-27 00:37:59 -07:00 |
|
test_cross_entropy_reused_logits.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_cross_entropy_softcap_padding.py
|
Name the Ascend NPU as its own tests expect, and close the spoofable skip-guard hole (#10968)
|
2026-09-14 23:07:25 -07:00 |
|
test_cuda_spoof_reports_free_memory.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_cuda_spoof_supplies_the_raw_stream.py
|
CUDA spoof: cover CUDA wheels on GPU-less runners, not only CPU wheels (#14056)
|
2026-10-17 01:23:58 -07:00 |
|
test_cudnn_sdpa_d256_masked_backward.py
|
Keep masked head_dim 256 SDPA training off cuDNN attention on SM100 (torch 2.14 NaN grads) (#12344)
|
2026-09-30 10:45:40 -07:00 |
|
test_dataclass_default_backfill.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_datasets_audio_fallback.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_ddp_batched_grad_copy.py
|
Opt-in DDP batched_grad_copy (UNSLOTH_DDP_BATCHED_GRAD_COPY=1) (#13350)
|
2026-10-12 05:09:10 -07:00 |
|
test_ddp_find_unused_default.py
|
DDP LoRA: default find_unused_parameters=False so gradient checkpointing does not crash (#13267)
|
2026-10-13 11:49:13 -07:00 |
|
test_ddp_rank_dispatch_hooks.py
|
Fix 4-bit VLM DDP training crash on ranks >= 1 (CheckpointError) (#13149)
|
2026-10-09 19:44:12 -07:00 |
|
test_debug_distributed_env.py
|
Add unsloth doctor --gpus multi-GPU preflight and UNSLOTH_DEBUG_DISTRIBUTED (#13884)
|
2026-10-16 03:49:31 -07:00 |
|
test_decision_clef_head.py
|
Fix DecisionTrainer crashing on step 1 without a prior inference pass (#13176)
|
2026-10-09 21:26:19 -07:00 |
|
test_decision_datasets.py
|
Decision models: follow-up fixes after #12772, #12876, #12939 (#12949)
|
2026-10-07 09:06:56 -07:00 |
|
test_decision_from_lm.py
|
Train any text or vision LLM as a Clef decision model from Studio, with FastDecisionModel.predict and adapter saves (#12876)
|
2026-10-07 05:52:14 -07:00 |
|
test_decision_gguf.py
|
tests: keep the Kaggle launcher's signal handlers out of the pytest worker (#12974)
|
2026-10-07 12:12:38 -07:00 |
|
test_decision_gguf_real.py
|
Fix the backend CI guards that main fails after the decision and ComfyUI merges (#12958)
|
2026-10-07 10:05:05 -07:00 |
|
test_decision_images.py
|
Fix image inputs in Clef training and prediction (#13471)
|
2026-10-13 04:55:17 -07:00 |
|
test_decision_mlx_guards.py
|
MLX: train decision models on Apple Silicon (#13016)
|
2026-10-08 12:04:53 -07:00 |
|
test_decision_model.py
|
Fix image inputs in Clef training and prediction (#13471)
|
2026-10-13 04:55:17 -07:00 |
|
test_decision_model_mlx.py
|
MLX: Clef Omni in Unsloth Studio, with audio in the Decision API and decision training (#13538)
|
2026-10-14 09:10:59 -07:00 |
|
test_declared_no_sdpa.py
|
Attention resolver: respect a declared _supports_sdpa = False, and skip flash_attention_2 when a class's compatible flash kernels exclude it (MiMo-V2-Flash) (#12147)
|
2026-09-28 01:33:05 -07:00 |
|
test_decode_gqa_in_place.py
|
Native decode: read GQA KV heads in place instead of copying them per query head (#13271)
|
2026-10-13 12:59:47 -07:00 |
|
test_deepseek_v41_non_reentrant_gc.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_deepspeed_zero3.py
|
Train with DeepSpeed ZeRO-2 and ZeRO-3 (#13587)
|
2026-10-14 04:12:07 -07:00 |
|
test_deliberate_crashes_suppress_cores.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_device_helpers.py
|
Build RoPE cos/sin caches only on the device that uses them (#13855)
|
2026-10-15 21:50:58 -07:00 |
|
test_dia_generate_pad_token.py
|
Run GRPO under FSDP2: no inference mode around FSDP, an optimizer that steps DTensors, and refuse vLLM with FSDP2 (#13156)
|
2026-10-11 04:22:20 -07:00 |
|
test_dill_module_by_value_fix.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_dispatch_hook_repair.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_empty_logits_compiled_path_probes.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_empty_logits_protocol_probes.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_enforce_kwargs_spacing.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_enforce_kwargs_spacing_mode.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_eval_mode_backward.py
|
Allow backward through eval-mode and for_inference forwards (#13050)
|
2026-10-08 07:29:31 -07:00 |
|
test_eval_step_restores_return_logits.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_eval_steps_ignored_warning.py
|
refactor(trainer): add warning for ignored eval_steps (#4225)
|
2026-09-29 07:02:53 -07:00 |
|
test_fa2_fast_generate_bypass.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fast_gemv_dispatch.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fast_generate_slow_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fast_inference_bnb_fallback.py
|
fast_inference: Gemma 4 Unified, Muse Glimmer, dense Gemma 4; 4-bit fallback when no -bnb-4bit repo (#13825)
|
2026-10-15 07:23:51 -07:00 |
|
test_fast_inference_old_gpu_fallback.py
|
fast_inference: switch Gemma 4 to Unsloth inference on GPUs below sm80 (#13864)
|
2026-10-15 08:41:15 -07:00 |
|
test_fast_inference_refuses_fsdp2.py
|
Run GRPO under FSDP2: no inference mode around FSDP, an optimizer that steps DTensors, and refuse vLLM with FSDP2 (#13156)
|
2026-10-11 04:22:20 -07:00 |
|
test_fast_language_model_after_fast_model.py
|
Let FastModel and FastLanguageModel load the same family in one process, in either order (#12219)
|
2026-09-29 12:17:06 -07:00 |
|
test_fastmodel_new_tokens_reload.py
|
Train embed_tokens and lm_head in FastModel after add_new_tokens (#13057)
|
2026-10-11 22:26:39 -07:00 |
|
test_fastmodel_prevolta_fast_inference.py
|
Nemotron-H fast_inference via vLLM weight sharing (fallback on older unsloth_zoo); drop vLLM-only kwargs on fallback (#13440)
|
2026-10-13 09:05:41 -07:00 |
|
test_finetune_last_n_layers.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fla_autotuner_fast_path.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flash_attention_mrope_packed.py
|
Keep flash attention from reading Qwen3.5 mRoPE position ids as packed sequences (#12856)
|
2026-10-07 05:06:44 -07:00 |
|
test_flash_attention_sub_config_fallback.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flash_attention_unregistered_tower.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flash_attn_4_namespace_shadow.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flash_compatible_implementations.py
|
Attention resolver: respect a declared _supports_sdpa = False, and skip flash_attention_2 when a class's compatible flash kernels exclude it (MiMo-V2-Flash) (#12147)
|
2026-09-28 01:33:05 -07:00 |
|
test_flex_attention_needs_ampere.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flex_gemma4_sliding_scope.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flex_head_dim_version_gate.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flex_large_head_dim_fallback.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flex_large_head_dim_kernel_options.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_flex_mask_graph_breaks.py
|
Build flex attention masks without graph breaks under torch.compile (#12837)
|
2026-10-06 05:43:58 -07:00 |
|
test_flex_maskless_sdpa.py
|
Send maskless causal flex calls to SDPA is_causal (#12773)
|
2026-10-06 03:52:57 -07:00 |
|
test_float32_generate_autocast.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_float32_lora_fallback.py
|
Fall back to PEFT's LoRA forward for float32 base weights (#7020)
|
2026-09-29 04:49:10 -07:00 |
|
test_float32_no_fp16_autocast.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_for_training_gc_off.py
|
Keep the load-time gradient checkpointing mode in for_training (#12629)
|
2026-10-03 11:10:05 -07:00 |
|
test_for_training_generation_flag.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_formatter_fixed_point.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_compile_fullgraph.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_device_context.py
|
[XPU] Enable FP8 training on XPU (#12535)
|
2026-10-04 05:55:58 -07:00 |
|
test_fp8_device_gate.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_down_proj_bf16_backward.py
|
Fused LoRA kernels: support lora_dropout and lora_bias (#13898)
|
2026-10-16 06:47:42 -07:00 |
|
test_fp8_fbgemm_block_shapes.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_int64_offsets.py
|
Fix illegal memory access in block-FP8 LoRA kernels on long batches (#13141)
|
2026-10-10 00:59:41 -07:00 |
|
test_fp8_leftover_dequant_16bit.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_moe_fast_inference_gate.py
|
FP8 MoE + fast_inference: refuse unsupported checkpoints before vLLM loads, with a hint that works (#13277)
|
2026-10-13 12:52:59 -07:00 |
|
test_fp8_pre_sm89_dequant_fallback.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_ragged_block_dequant.py
|
Dequantize block-FP8 weights with a ragged last block on 16-bit loads (GLM-5.3) (#12133)
|
2026-09-28 01:54:13 -07:00 |
|
test_fp8_restore_dropped_scale.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_rowwise_backend.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_rowwise_gemv_blackwell_gate.py
|
FP8 GEMV on Blackwell + Triton 3.3: no graph break under torch 2.7, keep the exact GEMV shapes that are faster (#14026)
|
2026-10-16 10:52:12 -07:00 |
|
test_fp8_rowwise_gemv_long_indexing.py
|
FP8 rowwise GEMV: int64 row offsets past 2^31 elements, and skip it on Blackwell with Triton 3.3 (#13830)
|
2026-10-15 09:55:17 -07:00 |
|
test_fp8_tiny_e8m0.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_to_nf4_load.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fp8_transposed_weight.py
|
[XPU] Enable FP8 training on XPU (#12535)
|
2026-10-04 05:55:58 -07:00 |
|
test_fp8_unscaled_checkpoint_linears.py
|
Keep Linear layers an FP8 checkpoint stores in bf16 unconverted (#12124)
|
2026-09-28 02:02:05 -07:00 |
|
test_fp8linear_high_precision_weight.py
|
[XPU] Enable FP8 training on XPU (#12535)
|
2026-10-04 05:55:58 -07:00 |
|
test_fp8linear_module_block_size.py
|
[XPU] Enable FP8 training on XPU (#12535)
|
2026-10-04 05:55:58 -07:00 |
|
test_frozen_norm_running_stats.py
|
Keep frozen BatchNorm running stats fixed during LoRA training (#12319)
|
2026-09-30 09:06:28 -07:00 |
|
test_fsdp2_config.py
|
Follow-ups to the GRPO reward locals and FSDP2 mixed precision merges (#14023)
|
2026-10-16 11:30:18 -07:00 |
|
test_fsdp2_default_optim.py
|
Run GRPO under FSDP2: no inference mode around FSDP, an optimizer that steps DTensors, and refuse vLLM with FSDP2 (#13156)
|
2026-10-11 04:22:20 -07:00 |
|
test_fsdp2_detection.py
|
FSDP2: first-class LoRA and full finetuning, keep fused LoRA kernels per block (#13865)
|
2026-10-15 23:21:43 -07:00 |
|
test_fsdp2_frozen_dtypes_gloo.py
|
FSDP2: first-class LoRA and full finetuning, keep fused LoRA kernels per block (#13865)
|
2026-10-15 23:21:43 -07:00 |
|
test_fsdp2_peft_checkpoint_gloo.py
|
FSDP2: first-class LoRA and full finetuning, keep fused LoRA kernels per block (#13865)
|
2026-10-15 23:21:43 -07:00 |
|
test_fsdp2_qlora.py
|
FSDP2: shard QLoRA 4-bit weights with fused LoRA kernels kept on (#13880)
|
2026-10-16 03:48:23 -07:00 |
|
test_fsdp2_qlora_prepare_order.py
|
FSDP2: shard QLoRA 4-bit weights with fused LoRA kernels kept on (#13880)
|
2026-10-16 03:48:23 -07:00 |
|
test_fsdp2_save_args.py
|
FSDP2: save_pretrained, merged saves and 4-bit checkpoints for sharded models (#13895)
|
2026-10-16 03:49:04 -07:00 |
|
test_fsdp2_save_gloo.py
|
FSDP2: save_pretrained, merged saves and 4-bit checkpoints for sharded models (#13895)
|
2026-10-16 03:49:04 -07:00 |
|
test_fsdp2_wrap_gloo.py
|
FSDP2: first-class LoRA and full finetuning, keep fused LoRA kernels per block (#13865)
|
2026-10-15 23:21:43 -07:00 |
|
test_fsdp_declines_fused_lora.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fsdp_generation_grad_mode.py
|
Run GRPO under FSDP2: no inference mode around FSDP, an optimizer that steps DTensors, and refuse vLLM with FSDP2 (#13156)
|
2026-10-11 04:22:20 -07:00 |
|
test_full_eval_keeps_training_state.py
|
Keep the training state intact across a standalone evaluate() / predict() (#11860)
|
2026-09-25 13:27:36 -07:00 |
|
test_full_finetune_prequantized_bnb.py
|
Dequantize pre-quantized bitsandbytes checkpoints for full finetuning (#13079)
|
2026-10-08 11:48:05 -07:00 |
|
test_fused_ce_logit_scale.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fused_ce_not_return_dict_logits.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_fused_lora_dropout_bias.py
|
Fused LoRA kernels: support lora_dropout and lora_bias (#13898)
|
2026-10-16 06:47:42 -07:00 |
|
test_fused_preference_loss.py
|
DPO, ORPO, CPO: opt-in fused preference loss (UNSLOTH_FUSED_PREFERENCE_LOSS=1) (#13870)
|
2026-10-15 22:18:13 -07:00 |
|
test_gc_train_scope.py
|
Freeze the setup-time heap during Trainer.train(); add resume parity and memory-flat GPU tests (#13332)
|
2026-10-13 10:44:38 -07:00 |
|
test_gemma2_attention_masks.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gemma2_flash_decode.py
|
Gemma2: use flash_attn_with_kvcache for cached decoding (#12112)
|
2026-09-27 22:54:22 -07:00 |
|
test_gemma2_softcap_int32_split.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gemma4_chat_template.py
|
Let the Gemma-4 chat template format tool-calling conversations (#14015)
|
2026-10-18 05:42:48 -07:00 |
|
test_gemma4_empty_thought.py
|
Let the Gemma-4 chat template format tool-calling conversations (#14015)
|
2026-10-18 05:42:48 -07:00 |
|
test_gemma4_split_vision_mask.py
|
Let Gemma 4 31B train when it is split across GPUs (#12233)
|
2026-09-29 13:03:10 -07:00 |
|
test_gemma_2b_mapper_key.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gemma_embed_scale.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gemma_fast_decode_float32.py
|
Copy the input in the Gemma inference RMSNorm so float32 decode keeps its residual (#13643)
|
2026-10-14 10:08:03 -07:00 |
|
test_gemma_multimodal_cache_selection.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_generate_kwarg_gate.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_generation_failure_is_visible.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_get_chat_template_processor.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_get_model_name.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gguf_basename_platform_matrix.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gguf_disk_headroom.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gguf_model_basename.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gguf_windows_export_routing.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gguf_windows_native.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gptq_packed_lora.py
|
Let AWQ checkpoints train with LoRA (#13584)
|
2026-10-13 04:37:16 -07:00 |
|
test_gradient_checkpointing_reentrant.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_gradient_checkpointing_restore.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_granite_float16_attention_dropout.py
|
Turn off Granite 3.0-3.2 attention dropout on float16 loads (#13417)
|
2026-10-12 21:56:20 -07:00 |
|
test_grouped_gemm_compile_and_int64.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grouped_gemm_flatten.py
|
MoE grouped GEMM: tl.range for the dX expert loop, drop the dead dW FLATTEN (#13831)
|
2026-10-15 10:17:53 -07:00 |
|
test_grouped_gemm_frozen_weight_grad.py
|
Kernel follow-ups: norm launch on HIP and under compile, grouped GEMM frozen operands, no CE eviction hint on H100 (#13967)
|
2026-10-16 09:00:41 -07:00 |
|
test_grouped_gemm_optional_gather_indices.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grouped_linear_lora.py
|
Trim comments in the DeepSeek-V4 grouped LoRA, remote-code shims and Mistral-format loader code (#12089)
|
2026-09-27 06:10:14 -07:00 |
|
test_growth_ratio_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_accumulated_loss_hidden_states_signal.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_autocast_disabled.py
|
GRPO: latch autocast under vLLM only, zero-variance exclusion on TRL before 1.6 (#14123)
|
2026-10-18 17:51:30 -07:00 |
|
test_grpo_autocast_per_trainer.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_clip_ratio_one_token.py
|
GRPO: exclude masked rows from clip_ratio metrics for one-token completions (#13883)
|
2026-10-16 17:52:46 +13:00 |
|
test_grpo_ddp_average_gradients_foreach.py
|
GRPO DDP gradient sync: copy each averaged bucket back with one _foreach_copy_ (#13268)
|
2026-10-13 12:59:43 -07:00 |
|
test_grpo_efficient_autocast_latch.py
|
GRPO: latch autocast under vLLM only, zero-variance exclusion on TRL before 1.6 (#14123)
|
2026-10-18 17:51:30 -07:00 |
|
test_grpo_eval_accumulation.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_eval_mode_restored_for_loss.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_fsdp2_scoring_head.py
|
Read the current FSDP2 output head when scoring GRPO log-probs (#14060)
|
2026-10-18 08:28:22 -07:00 |
|
test_grpo_hidden_states_logits_cost.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_hidden_states_per_call_degradation.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_hidden_states_signal.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_hidden_states_wrap_target.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_image_column_normalisation.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_is_metrics_mask.py
|
Support TRL 1.14 and 1.15: ORPO/CPO row cap, masked GRPO sampling metrics, cap to 1.15.0 (#13203)
|
2026-10-10 07:20:52 -07:00 |
|
test_grpo_legacy_trl_vision.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_move_model_to_vllm_sync.py
|
GRPO: keep TRL vLLM weight sync on TRL < 0.28 unless fast_inference shares weights (#13911)
|
2026-10-16 02:25:14 -07:00 |
|
test_grpo_old_logprob_gate.py
|
GRPO: keep TRL's old logprob gate when using vLLM (#13856)
|
2026-10-15 20:26:27 -07:00 |
|
test_grpo_packed_nograd_oom_recovery.py
|
GRPO no-grad packing: check the OOM length cap before flattening ids (#13980)
|
2026-10-16 03:08:44 -07:00 |
|
test_grpo_packed_raw_logits_nograd.py
|
GRPO packed logprobs: an OOM caps the packed length instead of disabling packing (#13877)
|
2026-10-15 21:25:10 -07:00 |
|
test_grpo_padded_raw_logits.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_park_optimizer_during_rollout.py
|
GRPO: opt-in parking of optimizer state and gradients on host during rollout (#13866)
|
2026-10-15 20:29:43 -07:00 |
|
test_grpo_ref_model_output_head.py
|
Forward output head bias in GRPO policy scoring (#13302)
|
2026-10-12 01:54:03 -07:00 |
|
test_grpo_reward_locals_cycle.py
|
GRPO: latch autocast under vLLM only, zero-variance exclusion on TRL before 1.6 (#14123)
|
2026-10-18 17:51:30 -07:00 |
|
test_grpo_sample_weight.py
|
GRPO: latch autocast under vLLM only, zero-variance exclusion on TRL before 1.6 (#14123)
|
2026-10-18 17:51:30 -07:00 |
|
test_grpo_video_inputs.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_vision_kwargs_forwarded.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_vision_reward_inputs.py
|
Let TRL AsyncGRPOTrainer train Unsloth models; fix GRPO vision rewards for tool images and TRL 1.15 (#13205)
|
2026-10-10 07:22:58 -07:00 |
|
test_grpo_vision_zoo_floor.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_grpo_width_dispatch_sites.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_harmony_tool_call_eos_5162.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_heterogeneous_config_probes.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_hf_login_keeps_saved_token.py
|
Keep the saved Hugging Face login when from_pretrained gets a token (#13486)
|
2026-10-13 17:38:23 +13:00 |
|
test_hf_packed_attention.py
|
Gemma 4 packed SFT: efficient kernel for the 512-wide global layers, processor on text-only data (#13384)
|
2026-10-12 06:30:56 -07:00 |
|
test_hf_xet_broken_wheel.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_high_precision_layernorm_coverage.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_idefics3_fast_inference_registration.py
|
Add Idefics3 support (Granite Docling VLM) (#4241)
|
2026-09-28 08:40:35 -07:00 |
|
test_ignored_tokenizer_casing.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_image_processing_reexports.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_import_fixes_drift.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_import_time_floors.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_inputs_embeds_require_grads.py
|
Make precomputed inputs_embeds train under gradient checkpointing (#13088)
|
2026-10-09 05:05:53 -07:00 |
|
test_installer_elevation_report.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_installer_interactive_prompts.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_installer_profile_hardening.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_installer_shortcut_icons.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_installer_skip_autostart.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_installer_system32_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_installer_systemd.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_installer_unsloth_version.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_installer_zoo_floor_parity.py
|
Studio: install from a package index that lags PyPI (#14133)
|
2026-10-18 09:50:33 -07:00 |
|
test_k3_mxfp4_stacked_experts.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_kaggle_gguf_error_message.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_kbit_prep_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_kimi_medias_processor_loader.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_large_head_attention.py
|
head_dim 512 attention: sm80+ only, no 256 MB autotune buffer, OOM falls back to SDPA, no per-call mask sync (#13977)
|
2026-10-16 04:17:34 -07:00 |
|
test_layernorm_gradient_layout.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_layernorm_precision_env_scope.py
|
Scope UNSLOTH_HIGH_PRECISION_LAYERNORM to the load that sets it (#12873)
|
2026-10-07 05:01:07 -07:00 |
|
test_layernorm_weight_grad.py
|
Kernel follow-ups: NVFP4 shape-aware fused gate, MoE fallback tiles, test fixes across torch 2.10 - 2.14 (#14122)
|
2026-10-18 08:22:25 -07:00 |
|
test_legacy_config_type_coercion.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_lint_duplicate_definitions.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_lint_no_parallel_clamp.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_llama_left_padding_training_mask.py
|
Keep the padding mask for left-padded label-less training forwards (#13154)
|
2026-10-11 00:11:53 -07:00 |
|
test_llava_num_additional_image_tokens.py
|
Fix LLaVA image token count when processor_config lacks num_additional_image_tokens (#13091)
|
2026-10-09 04:41:46 -07:00 |
|
test_loader_glob_skip.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_longcat_lsa.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_longrope_short_long_boundary.py
|
Build RoPE cos/sin caches only on the device that uses them (#13855)
|
2026-10-15 21:50:58 -07:00 |
|
test_lora_dropout_patch_summary.py
|
Fused LoRA kernels: support lora_dropout and lora_bias (#13898)
|
2026-10-16 06:47:42 -07:00 |
|
test_lora_gradient_layout.py
|
fix: reshape noncontiguous gradients in fused LoRA backward (#10712)
|
2026-10-12 08:19:33 -07:00 |
|
test_lora_init.py
|
Save LoRA-GA adapters relative to the original base so they reload correctly (#13322)
|
2026-10-11 20:30:37 -07:00 |
|
test_lora_traced_inplace.py
|
Fused Triton NF4 dequant and GEMV: bit-exact, stream-safe, torch.compile traceable (#12113)
|
2026-10-02 06:32:07 -07:00 |
|
test_managed_tools_master_root.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_map_eos_token.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_mapper_no_duplicate_keys.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_mapper_preserves_repo_id_case.py
|
Preserve repo id case when resolving a model name (#8058)
|
2026-09-29 13:06:29 -07:00 |
|
test_memory_flat_gpu.py
|
Freeze the setup-time heap during Trainer.train(); add resume parity and memory-flat GPU tests (#13332)
|
2026-10-13 10:44:38 -07:00 |
|
test_merged_hub_destination.py
|
Fine-tune and export native MTP heads and DSpark drafters (#13719)
|
2026-10-15 08:35:25 -07:00 |
|
test_mfu_logging.py
|
Add opt-in TFLOPs and MFU logging with LoRA aware FLOP counts (#13885)
|
2026-10-16 05:32:47 -07:00 |
|
test_mha_out_proj_bnb_dequant.py
|
Dequantize 4-bit nn.MultiheadAttention out_proj after load (#13090)
|
2026-10-09 05:13:25 -07:00 |
|
test_missing_optional_dep_skips.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_missing_torchvision_vlm.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_mistral_format_checkpoint.py
|
Trim comments in the DeepSeek-V4 grouped LoRA, remote-code shims and Mistral-format loader code (#12089)
|
2026-09-27 06:10:14 -07:00 |
|
test_mistral_format_mistral4.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_mixed_nvidia_capabilities.py
|
Keep the plain capability cache on single-GPU hosts (#13962)
|
2026-10-16 00:22:47 -07:00 |
|
test_mllama_flash_attention_excluded.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_mm_packing.py
|
Padding-free rows for Gemma 4 image and audio SFT (#13385)
|
2026-10-12 06:51:56 -07:00 |
|
test_model_registry.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_modeling_only_config_fields.py
|
Refuse K-EXAONE 2.0 on a transformers that ignores its config (#12128)
|
2026-09-28 01:39:30 -07:00 |
|
test_modelopt_fp8_loading.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_moe_fallback_tile_dims.py
|
Kernel follow-ups: NVFP4 shape-aware fused gate, MoE fallback tiles, test fixes across torch 2.10 - 2.14 (#14122)
|
2026-10-18 08:22:25 -07:00 |
|
test_moe_fast_inference_gate.py
|
fast_inference: Gemma 4 Unified, Muse Glimmer, dense Gemma 4; 4-bit fallback when no -bnb-4bit repo (#13825)
|
2026-10-15 07:23:51 -07:00 |
|
test_moe_lora_targets.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_multi_image_grpo_chunking.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_multiple_active_lora_adapters.py
|
Skip the fused LoRA kernels when a DoRA adapter is active (#13254)
|
2026-10-11 01:00:50 -07:00 |
|
test_mxfp4_compressed_linear.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_mxfp4_lora_packed_experts_dequantize.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_nemotron_h_user_bnb_skip.py
|
Keep Nemotron-H mixer.out_proj unquantized under a caller's BitsAndBytesConfig (#12131)
|
2026-09-28 01:33:53 -07:00 |
|
test_new_mapper_fetched_fp8.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_new_mapper_no_global_leak.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_nf4_triton_dequant.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_nf4_triton_gemv.py
|
NF4 GEMV: keep the table load on 256 wide tiles (#14027)
|
2026-10-16 11:05:05 -07:00 |
|
test_no_placement_params_device_map.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_non_lora_peft_inference.py
|
Run non-LoRA PEFT adapters through their own forward in fast inference (#13578)
|
2026-10-13 04:42:19 -07:00 |
|
test_nonreentrant_gc_side_stream_offload.py
|
Use side stream offload hooks for non-reentrant gradient checkpointing (#13872)
|
2026-10-15 20:28:54 -07:00 |
|
test_norm_backward_grad_alias.py
|
RMSNorm: keep the final norm's own dX buffer when hidden states are returned to the caller (#14139)
|
2026-10-18 15:25:20 -07:00 |
|
test_norm_fast_launch.py
|
Kernel follow-ups: NVFP4 shape-aware fused gate, MoE fallback tiles, test fixes across torch 2.10 - 2.14 (#14122)
|
2026-10-18 08:22:25 -07:00 |
|
test_norm_modules_to_save.py
|
Allow norm layers in FastLanguageModel modules_to_save (#13805)
|
2026-10-15 04:31:48 -07:00 |
|
test_nvfp4_fused_gate_shape.py
|
Kernel follow-ups: NVFP4 shape-aware fused gate, MoE fallback tiles, test fixes across torch 2.10 - 2.14 (#14122)
|
2026-10-18 08:22:25 -07:00 |
|
test_nvfp4_fused_gemm.py
|
NVFP4 tests: real-CUDA skip guards, and format the files #13840 left off the hook (#13937)
|
2026-10-15 14:37:02 -07:00 |
|
test_nvfp4_kernel.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_nvfp4_loader.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_nvfp4_lora.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_nvfp4_quant_load.py
|
Trim comments in tests, scripts, CI helpers and unsloth_cli (#10113)
|
2026-09-04 15:31:50 -07:00 |
|
test_nvidia_smi_torch_mismatch_hint.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offline_loading_helpers.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offload_auto_trainer_replan.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offload_device_property.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offload_embedding_hooks.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offload_embedding_op_forward.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offload_embedding_trainer_move.py
|
Keep offloaded embeddings in host RAM when Trainer and accelerate place the model (#13320)
|
2026-10-11 16:13:27 -07:00 |
|
test_offload_frozen_module_device.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offload_tied_autodisable.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offload_tied_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_offloaded_parameter_hint.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_ollama_eos_token_order.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_omni_auto_class_resolution.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_omni_text_only_load.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_omni_trains_through_thinker.py
|
Hand a composition with no forward of its own to its thinker (Qwen3-Omni) (#11523)
|
2026-09-25 13:27:16 -07:00 |
|
test_online_dpo_left_padding.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_optimized_precision_conflicts.py
|
Merge main into #10280: one unsloth doctor for both checks
|
2026-10-18 08:33:13 -07:00 |
|
test_optimizer_state_offload.py
|
Optimizer state offload: keep inactive state on the host for unchunked Q-GaLore weight_quant steps (#14025)
|
2026-10-16 10:30:38 -07:00 |
|
test_padding_free_eval.py
|
Fix evaluate() under auto padding-free and keep compute_metrics rows per example (#13136)
|
2026-10-09 09:38:22 -07:00 |
|
test_padding_free_signature_gate.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_padding_free_train_use_cache.py
|
Never train with a KV cache, so padding-free rows keep their document boundaries (#13177)
|
2026-10-11 22:12:07 -07:00 |
|
test_past_kv_models.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_past_kv_utils.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_peft_layers_pattern.py
|
Pass layers_pattern from FastLanguageModel.get_peft_model to LoraConfig (#13751)
|
2026-10-15 07:03:14 -07:00 |
|
test_peft_moe_linear_targets.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_peft_stale_torchao.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_peft_str_target_modules.py
|
Pass str target_modules (regex or "all-linear") to PEFT on the Llama path (#13567)
|
2026-10-13 04:37:49 -07:00 |
|
test_peft_symbol_backfill.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_peft_tensor_parallel_compat.py
|
|
|
|
test_peft_torchao_missing_tensor_subclass.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_peft_weight_converter_compat.py
|
|
|
|
test_per_layer_device_readers.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_plain_load_after_unsloth_rope.py
|
Build RoPE cos/sin caches only on the device that uses them (#13855)
|
2026-10-15 21:50:58 -07:00 |
|
test_ppo_trainer_patches.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_pre_volta_compile_disable.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_pre_volta_triton_codegen.py
|
Run the RoPE and NF4 Triton kernels on GPUs older than Volta (#13153)
|
2026-10-11 01:21:50 -07:00 |
|
test_prefetch_snapshot_scope.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_prequantized_vlm_quant_state_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_pretrain_compile_reset.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_processor_text_packing_gate.py
|
Pack Qwen3.5 text SFT with the processor, and fix eval_loss on packed text runs (#13624)
|
2026-10-14 03:17:12 -07:00 |
|
test_profile_startup_gate.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_psutil_apple_cpu_freq.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_public_api_surface.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_push_to_ollama_modelfile_args.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_python39_compatibility.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_pythonpath_empty_components.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_qat_activation_fused_lora.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_qat_forward_fake_quant_fallback.py
|
QAT: fall back to the module forward for torchao NVFP4 / MX linears in fused LoRA (#13280)
|
2026-10-13 12:28:25 -07:00 |
|
test_qat_lm_head_fused_ce.py
|
Fix LM-head weight fake quantization in fused CE (#14006)
|
2026-10-18 09:16:46 -07:00 |
|
test_qk_norm_rope.py
|
Fused Qwen3 q_norm / k_norm RMSNorm + RoPE Triton kernel (opt-in) (#13890)
|
2026-10-16 00:31:27 -07:00 |
|
test_qlora_notice_with_quantization_config.py
|
Do not print the 16bit LoRA notice when a quantization_config still quantizes (#12848)
|
2026-10-06 08:38:02 -07:00 |
|
test_raw_text.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_raw_text_json_loading.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_rdna1_triton_buffer_ops.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_reject_alora.py
|
Reject Activated LoRA adapters instead of training them as plain LoRA (#13089)
|
2026-10-08 14:48:14 -07:00 |
|
test_reload_modules_to_save_trainable.py
|
Keep modules_to_save trainable when reloading a LoRA checkpoint (#13303)
|
2026-10-11 21:20:38 -07:00 |
|
test_remote_code_class_and_expert_submodules.py
|
Fix main's red CI after #13806, #13899 and recent formatter drift (#13909)
|
2026-10-15 11:29:52 -07:00 |
|
test_remote_code_dtype_cast_on_quantized_model.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_remote_code_input_embed_layer.py
|
Train EXAONE 3.5: name the token embedding remote code no longer exposes (#13058)
|
2026-10-08 08:18:24 -07:00 |
|
test_remote_code_model_api.py
|
Train Kimi-K3 with its MXFP4 experts kept packed and dequantized on the fly (#11750)
|
2026-09-28 12:25:51 -07:00 |
|
test_remote_code_non_persistent_buffers.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_remote_code_padded_head_params.py
|
Narrow Kimi-K3's zero-padded KDA A_log to num_heads so the fla backward runs (#12132)
|
2026-09-28 02:04:51 -07:00 |
|
test_remote_code_shims.py
|
Trim comments in the DeepSeek-V4 grouped LoRA, remote-code shims and Mistral-format loader code (#12089)
|
2026-09-27 06:10:14 -07:00 |
|
test_remote_code_slow_tokenizer_conversion.py
|
Keep slow tokenizers transformers cannot convert instead of crashing the load (#13073)
|
2026-10-09 04:37:05 -07:00 |
|
test_remote_code_training_gates.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_remote_code_transformers5_compat.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_remote_gradient_checkpointing.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_remote_moe_shims.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_remote_multimodal_cache_view.py
|
Phi-4-reasoning-vision: give remote multimodal prep an indexable cache view on transformers 5 (#12123)
|
2026-09-28 01:34:55 -07:00 |
|
test_remote_rope_scaling_none.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_repo_code_vlm_processor.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_resolve_model_class.py
|
|
|
|
test_resume_parity_gpu.py
|
Freeze the setup-time heap during Trainer.train(); add resume parity and memory-flat GPU tests (#13332)
|
2026-10-13 10:44:38 -07:00 |
|
test_return_logits_survives_for_training.py
|
Keep a user's UNSLOTH_RETURN_LOGITS=1 after for_inference() on FastModel (#13406)
|
2026-10-12 09:02:31 -07:00 |
|
test_rl_config_compat.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_rl_list_train_dataset.py
|
Accept a list train_dataset again on TRL 1.10+ (fixes vision notebooks) (#12628)
|
2026-10-03 14:38:06 -07:00 |
|
test_rms_layernorm_multirow.py
|
Run narrow RMSNorm rows several per program, bit-identical (Qwen3 q/k norms) (#12754)
|
2026-10-05 23:41:49 -07:00 |
|
test_rms_rope_compile.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_rmsnorm_gradient_layout.py
|
Name the Ascend NPU as its own tests expect, and close the spoofable skip-guard hole (#10968)
|
2026-09-14 23:07:25 -07:00 |
|
test_rmsnorm_save_output.py
|
RMSNorm save output: skip inference tensor weights, keep the default path free of opt-in checks (#13981)
|
2026-10-16 02:47:03 -07:00 |
|
test_rmsnorm_weight_grad.py
|
Compute RMSNorm weight gradient when the norm weight is trainable (#13594)
|
2026-10-14 09:41:32 -07:00 |
|
test_rocm_windows_fused_sdpa.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_root_hook_follows_input_embedding.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_rope_cache_current_device.py
|
Cover explicit RoPE position ids and compute the starting cache on CPU (#13976)
|
2026-10-16 04:47:37 -07:00 |
|
test_rope_sync_single_device.py
|
Skip the per-layer RoPE stream sync unless the model spans several devices (#13609)
|
2026-10-14 04:11:19 -07:00 |
|
test_rope_zero_stride_grad.py
|
Fix int32 offset overflow past 2^31 elements in RoPE and norm kernels (#13121)
|
2026-10-09 21:06:48 -07:00 |
|
test_run_ruff_format_pin.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_runtime_text_encoding.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_save_entrypoints_reach_converter.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_save_keep_fp32_modules.py
|
Keep fp32-only tensors on 16-bit merged saves (#13819)
|
2026-10-15 07:27:38 -07:00 |
|
test_save_lora_without_vllm.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_save_mtp_head_stitch.py
|
Fine-tune and export native MTP heads and DSpark drafters (#13719)
|
2026-10-15 08:35:25 -07:00 |
|
test_save_mtp_policy.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_save_tokenizer_class.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_scripts_write_lf.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_sdpa_fully_masked_rows.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_seq2seq_auto_class.py
|
Decide accepts_loss_kwargs from the loss head so gradient accumulation gets num_items_in_batch for every model (#12946)
|
2026-10-09 09:04:34 -07:00 |
|
test_settle_eager_fallbacks_between_steps.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_setup_master_root.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_sft_assistant_mask_fallback.py
|
assistant_only_loss with Unsloth chat templates: fall back to response markers (#13204)
|
2026-10-10 07:22:15 -07:00 |
|
test_sft_config_subclass_downcast.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_sft_vision_dataset_gate.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_source_read_encoding.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_speech_to_text_auto_class.py
|
Load speech-to-text models (Voxtral, Qwen2-Audio) through FastModel (#12127)
|
2026-09-28 02:03:03 -07:00 |
|
test_st_modules_json_trust_gate.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_st_save_merged_signature.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_st_subfolder_weights_are_fetched.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_st_transformer_subclass_reuse.py
|
FastSentenceTransformer: reuse the loaded model for Transformer subclass modules (#13591)
|
2026-10-13 07:47:44 -07:00 |
|
test_stale_kernel_hint.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_static_cache_memory_fallback.py
|
Fall back to a dynamic KV cache when the static one would not fit (#13075)
|
2026-10-10 06:11:41 -07:00 |
|
test_stopping_criteria_device.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_studio_install_workspace_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_studio_root_resilience.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_studio_shutdown_thread_wait.py
|
|
|
|
test_switch_training_mode.py
|
Switch between training and inference mode in one tree walk (#13374)
|
2026-10-12 03:34:38 -07:00 |
|
test_sync_load_when_quantizing.py
|
Load one tensor at a time when quantizing a 16-bit checkpoint on the fly (#12997)
|
2026-10-08 03:44:39 -07:00 |
|
test_synthetic_chunk_data.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_synthetic_hf_transfer_gate.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_synthetic_vllm_port.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_synthetic_vllm_startup_failure.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_temporary_patch_coverage.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_text_only_device_plan.py
|
Plan a device map for text_only loads of vision-language models (#11584)
|
2026-09-25 08:44:21 -07:00 |
|
test_text_only_parent_conversions.py
|
Load 4.x remote code and config-only remote code on transformers 5 (Trinity-Large, MiniMax-M3) (#11658)
|
2026-09-25 10:44:52 -07:00 |
|
test_text_only_prequantized_dtype.py
|
Cast fp16 leftovers to the requested dtype after a text_only pre-quantized load (#11692)
|
2026-09-25 07:54:03 -07:00 |
|
test_tiled_mlp_ddp.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_token_budget_batches.py
|
Opt-in token-balanced and token-budget SFT micro-batches (#13878)
|
2026-10-15 21:51:39 -07:00 |
|
test_token_budget_metrics.py
|
Token budget: weight split entropy by labels on TRL 1.3 to 1.14 (#14024)
|
2026-10-16 11:31:30 -07:00 |
|
test_tokenizer_bytelevel_roundtrip.py
|
Rebuild CohereTokenizer from tokenizer.json when transformers v5 changes its ids (#12130)
|
2026-09-28 01:56:41 -07:00 |
|
test_tokenizer_pickle.py
|
Make Unsloth-patched tokenizers and processors picklable (#13209)
|
2026-10-10 07:23:39 -07:00 |
|
test_tokenizers_v1.py
|
Add opt-in tokenizers 1.0 release candidate encode path (#13214)
|
2026-10-10 07:48:57 -07:00 |
|
test_tolerated_cast_and_compiled_packing.py
|
Load and train remote-code multimodal wrappers: Phi-4-reasoning-vision in 4-bit and Nemotron-3-Nano-Omni (#11526)
|
2026-09-25 12:16:36 -07:00 |
|
test_tool_mask_zoo_compat.py
|
|
|
|
test_torch2110_cuda_extras.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_torch_export_pt2_load_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_torch_getattr_warning_location.py
|
Keep torch deprecation warnings at the caller through the __getattr__ wrapper (#12635)
|
2026-10-03 13:41:02 -07:00 |
|
test_torch_without_distributed_backend.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_torchao_aten_grouped_mm.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_torchao_nf4tensor_move.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_torchao_nodist.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_torchao_subprocess_fix.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_torchao_torch_symbol_skew.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_torchaudio_cuda_mismatch.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_train_mode_asserted_once.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_trainer_data_parallel_guard.py
|
Fix notebook failures on Kaggle T4x2 and duplicate import warnings (#13024)
|
2026-10-08 04:13:01 -07:00 |
|
test_trainer_fp16_scaler_attach.py
|
Fix fp16 LoRA training with plain transformers.Trainer (fused CE gradients underflowed) (#13152)
|
2026-10-10 23:05:14 -07:00 |
|
test_transformers5_bare_annotation_live.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_transformers_cap_sites.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_transformers_dependency_floor.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_transformers_security_fixes.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_triton_fp_fusion.py
|
Fix main's red CI after #13806, #13899 and recent formatter drift (#13909)
|
2026-10-15 11:29:52 -07:00 |
|
test_triton_launch.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_triton_module_cache_key.py
|
Key triton < 3.4 driver module cache by Python version (#13409)
|
2026-10-12 07:17:54 -07:00 |
|
test_triton_optional_import.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_trl_cap_sites.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_trl_fused_lm_head_devices.py
|
Fix DPO on models split across GPUs with TRL 1.15's fused LM head (#13641)
|
2026-10-14 03:54:57 -07:00 |
|
test_trl_tool_chat_templates.py
|
AsyncGRPOTrainer with fast_inference: train and generate on one GPU with shared weights (#13292)
|
2026-10-11 20:11:01 -07:00 |
|
test_trust_remote_code_native_compile.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_typer_dependency_floor.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_ulysses.py
|
Context parallel: Ulysses backend with packing and varlen attention (#13848)
|
2026-10-15 21:15:20 -07:00 |
|
test_ulysses_hybrid.py
|
Context parallel: Ulysses for Gated DeltaNet hybrids (Qwen3.5, Qwen3-Next) (#13920)
|
2026-10-15 21:55:18 -07:00 |
|
test_uma_safetensors_load.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_uninitialized_position_ids.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_unpublished_16bit_mapping.py
|
Point five mapper rows at their upstream repo instead of an unpublished unsloth 16bit name (#12587)
|
2026-10-03 11:06:10 -07:00 |
|
test_unsloth_device_map_leaks.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_unsloth_device_map_optin.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_unsloth_device_map_platform_matrix.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_unsloth_device_map_review_fixes.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_unsloth_home_root_agreement.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_untrusted_vlm_native_class.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_upload_model_card_method.py
|
Name the method on the uploaded model card (#10816)
|
2026-09-13 19:57:40 -07:00 |
|
test_validate_rope_ignore_keys.py
|
Accept the transformers 5.0 ignore_keys argument in validate_rope (#11560)
|
2026-09-25 03:22:27 -07:00 |
|
test_version_single_source.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_video_path_validation.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_vllm_broken_detection.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_vllm_cuda_mismatch_wheel_url.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_vlm_processor_native_defaults.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_voxtral_training_fixes.py
|
Fix Voxtral processor saving and Voxtral Realtime training (#12647)
|
2026-10-14 04:01:41 -07:00 |
|
test_warnings_issued_guard.py
|
Add opt-in TFLOPs and MFU logging with LoRA aware FLOP counts (#13885)
|
2026-10-16 05:32:47 -07:00 |
|
test_wheel_smoke_publish_guard.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_windows_amd_gpu_scan_fallback.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_windows_no_sentencepiece.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_windows_rocm_bnb_version.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
|
test_xdist_torch_thread_cap.py
|
tests: give each xdist worker one CPU share for torch (#13494)
|
2026-10-12 18:24:55 -07:00 |
|
test_xformers_stub_without_metadata.py
|
Unsloth Studio (AMD/Windows): real torchao INT8 / FP8 export on Windows ROCm (#7102)
|
2026-09-28 13:54:10 -07:00 |
|
validate_studio_features.py
|
Trim comments in tests (#13844)
|
2026-10-15 05:30:27 -07:00 |
