ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
unsloth/unsloth
History
Exact
Daniel Han 55b49c8a1a Keep triton.enable_persistent_tma_matmul as >= 9, only gate CUTLASS options 
Triton TMA persistent matmul works on SM90+ (PyTorch's own
has_triton_tma_device() checks >= (9,0)). Only CUTLASS TMA-specific
kernels (cutlass_tma_only, cutlass_epilogue_fusion_enabled) are SM90-only
and need the == 9 gate.
2026-03-08 13:19:38 +00:00
..
dataprep Guard optional vLLM imports when extension is broken (#4068) 2026-02-15 22:09:29 -08:00
kernels fix(ROCm): restrict is_rdna() to ROCm-officially-supported RDNA GPUs (#4136) 2026-03-03 03:05:38 -08:00
models Keep triton.enable_persistent_tma_matmul as >= 9, only gate CUTLASS options 2026-03-08 13:19:38 +00:00
registry Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks" 2025-12-01 07:24:58 -08:00
utils Fix/pr 3699 leftpad prefill main (#4100) 2026-02-25 07:21:04 -08:00
__init__.py Fix broken wandb import crashing unsloth startup (#4147) 2026-03-03 07:08:12 -08:00
_auto_install.py Add PyTorch 2.10 and xformers 0.0.34 support (#3985) 2026-02-05 05:56:26 -08:00
chat_templates.py Fix regressions from security PRs #4042, #4044, and #4045 (#4062) 2026-02-15 23:16:17 -08:00
device_type.py Conditionally enable 4bit on CDNA for bitsandbytes>=v0.49.2 (#4161) 2026-03-07 01:33:40 -08:00
import_fixes.py Also patch accelerate's is_wandb_available for trl callbacks path (#4148) 2026-03-03 08:28:55 -08:00
ollama_template_mappers.py Refactor Ollama template wiring and harden packing helpers (#3890) 2026-02-09 04:04:48 -08:00
save.py fix: update GGUF save paths to use ~/.unsloth/llama.cpp with Windows support (#4138) 2026-03-03 06:34:09 -08:00
tokenizer_utils.py Add resilience to TRL internal API reclassification (#4111) 2026-02-25 06:34:21 -08:00
trainer.py Fix auto padding free logic to respect user passed False (#4128) 2026-03-01 19:30:47 -08:00