unsloth/unsloth
金黄色葡萄球君君 dd5ff9dcef ROCm: Add gfx950 (MI355X/CDNA4) to is_cdna() (#4051)
MI355X (gfx950) has the same 1024-thread workgroup limit as MI300X (gfx942),
but was missing from is_cdna(), causing all Triton kernels to use num_warps=32
(2048 threads) instead of 16 (1024 threads), resulting in OutOfResources crash.

Tested on: 8x AMD Instinct MI355X (gfx950), ROCm 7.1
2026-02-14 02:50:05 -08:00
..
dataprep [pre-commit.ci] auto fixes from pre-commit.com hooks 2026-01-08 11:35:21 +00:00
kernels ROCm: Add gfx950 (MI355X/CDNA4) to is_cdna() (#4051) 2026-02-14 02:50:05 -08:00
models Simplify MI300X startup banner name (#4049) 2026-02-14 02:24:03 -08:00
registry Revert "[pre-commit.ci] auto fixes from pre-commit.com hooks" 2025-12-01 07:24:58 -08:00
utils Refactor Ollama template wiring and harden packing helpers (#3890) 2026-02-09 04:04:48 -08:00
__init__.py Suppress HIP amdgpu.ids stderr noise during causal_conv1d check (#4052) 2026-02-14 02:44:34 -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 Refactor Ollama template wiring and harden packing helpers (#3890) 2026-02-09 04:04:48 -08:00
device_type.py Add missing import of inspect (#3778) 2025-12-25 18:43:59 -08:00
import_fixes.py Suppress HIP amdgpu.ids stderr noise during causal_conv1d check (#4052) 2026-02-14 02:44:34 -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 add llama.cpp prefix to gguf conversion help messages (#4016) 2026-02-10 01:59:05 -08:00
tokenizer_utils.py Fix notebook compatibility for transformers 4.57.6 and TRL 0.22-0.27 (#3998) 2026-02-09 05:11:50 -08:00
trainer.py Fix notebook compatibility for transformers 4.57.6 and TRL 0.22-0.27 (#3998) 2026-02-09 05:11:50 -08:00