12月1月2月3月4月5月6月7月8月9月10月11月
周一周三周五
一年内 261573 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/image-only-chat-validation 分支比较 2 提交 »61c23e5e7a test: deduplicate empty content validation coveraged6c39467dd fix: allow image-only chat messages
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/hip-amdgpu-ids-model-import-filter 分支cd0c708997 Apply amdgpu.ids fd2 filter during models import startup
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/gptoss-generate-poisons-compiled-training 分支比较 2 提交 »a1bd05d301 [pre-commit.ci] auto fixes from pre-commit.com hooks0fc33292c7 Reset torch.compile cache after a pre-train generate, not just a grad probe
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/gguf-order-test-inheritance 分支16e6bca247 tests: restore the inheritance-before-guard ordering assertion
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/gguf-audio 分支936c69585c Merge branch 'main' into fix/gguf-audio
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/gemma4-fp16-force-float32 分支06fbe97b54 Add Gemma-4 to FORCE_FLOAT32 list to fix fp16 NaN in RL training
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/gemma4-force-float32 分支比较 3 提交 »e264d42413 [pre-commit.ci] auto fixes from pre-commit.com hookse5ee16b8a1 Add FORCE_BFLOAT16 for Gemma-4 to prevent fp16 NaN in RL trainingd323b47805 Add Gemma-4 to FORCE_FLOAT32 to prevent fp16 NaN in RL training
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/fused-lora-fp32-dtype 分支比较 3 提交 »cb1ef59b52 Avoid recomputing LoRA B-projected gradient in fused-LoRA backwardf0705ef6e5 Tighten fused-LoRA dtype-fix comments08a985d28a Fix fused LoRA dtype mismatch on fp32 activations into fp16/bf16 weights
- frostbyte_neo 推送了仓库 github-featured/unslothai--unsloth 的 fix/fp8-moe-loader-guard 分支比较 2 提交 »418b43dca2 [pre-commit.ci] auto fixes from pre-commit.com hooksdde18ddd28 Fix FP8 MoE loader: guard import, pass auth to FastModel
