branch: master
Commits on master
- 55ac2a2 Tensor.cat with 0 shape tensors (#3062) 2 years ago
- f0d7ad8 fix gpt2 attention with start_pos = 0 (#3061) 2 years ago
- 39b9113 Speed tweaks (#3059) 2 years ago
- bf6281f hotfix: remove useless slow assert from ShapeTracker 2 years ago
- 4b687af explicit lazybuffer caching (#3058) 2 years ago
- 2c6f2e8 No extra vars call (#3054) 2 years ago
- 259bf9b add multigpu test for RMSNorm (#3056) 2 years ago
- dab8214 unit tests for Device.canonicalize (#3055) 2 years ago
- 374f765 remove unused reciprocal (#3053) 2 years ago
- a686663 make Embedding device aware for multigpu (#3051) 2 years ago
- 19298e7 Device._buffers -> Device._devices (#3052) 2 years ago
- 4f4e863 use in_features directly in nn.Linear.__init__ bound check (#3050) 2 years ago
- ee6a738 clean up test_nn.py (#3049) 2 years ago
- 3eb3664 fix nn.Embedding with empty length input (#3048) 2 years ago
- 7ea2e00 move children for speed (#3047) 2 years ago
- 655c6f6 St real size (#3046) 2 years ago
- 1d730b8 remove ACCUM_FP32 in simple_matmul.py (#3045) 2 years ago
- 47d67da track the size in the lazybuffer (#3044) 2 years ago
- c003be7 Revert "track size in shapetracker" (#3043) 2 years ago
- 50754f1 add caches there (#3042) 2 years ago
- c5a941d webgl backend in extra (#3041) 2 years ago
- 8cbcd1b Remove webgpu, back to 5k lines (#3040) 2 years ago
- cf2eea9 more beautiful_cartpole with exposed hparams 2 years ago
- 4461842 add bf16 type_map for both cuda and hip (#3036) 2 years ago
- ef5f545 add more Tensor.clip test cases (#3034) 2 years ago
- c9371f0 hotfix llama conversation mode (#3031) 2 years ago
- fa707c8 move beautiful cartpole action sampling inside jit (#3028) 2 years ago
- ebb81e8 hotfix: st.size() -> st.size in llama 2 years ago
- a8ba1ac track size in shapetracker (#3026) 2 years ago
- 138c17c enable argmax tests for METAL/WEBGPU in CI (#3027) 2 years ago