Preview

stanford-futuredata-megablocks-aa67416
- .gitignore
  
  73 Bytes
- .gitmodules
  
  132 Bytes
- Dockerfile
  
  239 Bytes
- LICENSE
  
  22.1 kB
- MANIFEST.in
  
  54 Bytes
- Makefile
  
  181 Bytes
- README.md
  
  2.8 kB
- csrc
  - cuda_util.h
    
    1.3 kB
  - cumsum.h
    
    4.6 kB
  - histogram.h
    
    2.7 kB
  - indices.h
    
    3.1 kB
  - ops.cu
    
    745 Bytes
  - replicate.h
    
    6.4 kB
  - sort.h
    
    2.7 kB
- docker.sh
  
  195 Bytes
- exp
  - dmoe
    - dmoe_125m_8gpu.sh
      
      3.4 kB
    - dmoe_356m_8gpu.sh
      
      3.4 kB
    - dmoe_46m_8gpu.sh
      
      3.4 kB
    - dmoe_760m_8gpu.sh
      
      3.4 kB
  - gpt2
    - gpt2_125m_1gpu.sh
      
      2.8 kB
    - gpt2_125m_8gpu.sh
      
      3.0 kB
    - gpt2_1315m_1gpu.sh
      
      2.8 kB
    - gpt2_1315m_8gpu.sh
      
      3.0 kB
    - gpt2_356m_1gpu.sh
      
      2.8 kB
    - gpt2_356m_8gpu.sh
      
      3.0 kB
    - gpt2_46m_1gpu.sh
      
      2.8 kB
    - gpt2_46m_8gpu.sh
      
      3.0 kB
    - gpt2_760m_1gpu.sh
      
      2.8 kB
    - gpt2_760m_8gpu.sh
      
      3.0 kB
  - moe
    - moe_125m_8gpu.sh
      
      3.4 kB
    - moe_356m_8gpu.sh
      
      3.4 kB
    - moe_46m_8gpu.sh
      
      3.4 kB
- media
  - dropping_end_to_end.png
    
    1.6 MB
- megablocks
  - __init__.py
    
    59 Bytes
  - backend
    - __init__.py
      
      1 Byte
    - kernels.py
      
      16.5 kB
  - benchmark_util.py
    
    817 Bytes
  - grouped_gemm_util.py
    
    488 Bytes
  - layers
    - __init__.py
      
      1 Byte
    - all_to_all.py
      
      1.4 kB
    - arguments.py
      
      2.5 kB
    - common.py
      
      606 Bytes
    - dmoe.py
      
      10.3 kB
    - dmoe_test.py
      
      6.6 kB
    - gelu.py
      
      1.2 kB
    - memory_test.py
      
      2.9 kB
    - memory_test.sh
      
      246 Bytes
    - mlp.py
      
      20.2 kB
    - moe.py
      
      16.8 kB
    - moe_test.py
      
      4.4 kB
    - mpu.py
      
      3.1 kB
    - parallelism_test.py
      
      4.5 kB
    - parallelism_test.sh
      
      251 Bytes
    - router.py
      
      2.2 kB
    - testing.py
      
      1.0 kB
    - weight_parallel.py
      
      11.4 kB
  - ops
    - __init__.py
      
      707 Bytes
    - all_to_all_benchmark.py
      
      1.5 kB
    - all_to_all_benchmark.sh
      
      252 Bytes
    - binned_gather.py
      
      736 Bytes
    - binned_gather_test.py
      
      2.0 kB
    - binned_scatter.py
      
      1.2 kB
    - binned_scatter_test.py
      
      2.4 kB
    - cumsum.py
      
      1.2 kB
    - cumsum_test.py
      
      1.1 kB
    - gather.py
      
      755 Bytes
    - histogram.py
      
      515 Bytes
    - histogram_benchmark.py
      
      2.1 kB
    - histogram_test.py
      
      2.4 kB
    - matmul_benchmark.py
      
      11.5 kB
    - padded_gather.py
      
      860 Bytes
    - padded_gather_test.py
      
      2.7 kB
    - padded_scatter.py
      
      2.7 kB
    - padded_scatter_benchmark.py
      
      1.7 kB
    - padded_scatter_test.py
      
      4.0 kB
    - permute_benchmark.py
      
      4.8 kB
    - repeat.py
      
      122 Bytes
    - replicate.py
      
      961 Bytes
    - replicate_test.py
      
      3.1 kB
    - round_up.py
      
      332 Bytes
    - scatter.py
      
      2.4 kB
    - sort.py
      
      775 Bytes
    - sort_benchmark.py
      
      2.2 kB
    - sort_test.py
      
      1.8 kB
    - sum.py
      
      120 Bytes
    - topology.py
      
      944 Bytes
    - topology_test.py
      
      2.3 kB
  - turbo_util.py
    
    483 Bytes
- requirements.txt
  
  177 Bytes
- setup.py
  
  1.5 kB
- third_party
  - Megatron-LM
- yamls
  - matmul_benchmark.yaml
    
    694 Bytes
  - triton_benchmark.yaml
    
    427 Bytes

stanford-futuredata/megablocks-v0.3.2.zip