Hugging Face's logo

aws-neuron
/
optimum-neuron-cache

Model card Files Community
208
dacorvo HF staff commited on
Commit
6319895
•
1 Parent(s): e9debb0

Synchronizing local compiler cache.

Browse files
Files changed (15) hide show
  1. .gitattributes +1 -0
  2. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.20/inference/llama/01-ai/Yi-9B/95ef5301db7f1b728496.json +1 -0
  3. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/gpt2/gpt2/3bcf76c897d15e9f8e43.json +1 -0
  4. neuronxcc-2.12.68.0+4480452af/MODULE_a4b5e92c38528bfc2597+2c2d707e/compile_flags.txt +1 -0
  5. neuronxcc-2.12.68.0+4480452af/MODULE_a4b5e92c38528bfc2597+2c2d707e/model.hlo.pb +3 -0
  6. neuronxcc-2.12.68.0+4480452af/MODULE_a4b5e92c38528bfc2597+2c2d707e/model.neff +0 -0
  7. neuronxcc-2.12.68.0+4480452af/MODULE_c92166730ec03623bbc8+2c2d707e/compile_flags.txt +1 -0
  8. neuronxcc-2.12.68.0+4480452af/MODULE_c92166730ec03623bbc8+2c2d707e/model.hlo.pb +3 -0
  9. neuronxcc-2.12.68.0+4480452af/MODULE_c92166730ec03623bbc8+2c2d707e/model.neff +0 -0
  10. neuronxcc-2.12.68.0+4480452af/MODULE_ce6900904f4375e48d21+2c2d707e/compile_flags.txt +1 -0
  11. neuronxcc-2.12.68.0+4480452af/MODULE_ce6900904f4375e48d21+2c2d707e/model.hlo.pb +3 -0
  12. neuronxcc-2.12.68.0+4480452af/MODULE_ce6900904f4375e48d21+2c2d707e/model.neff +0 -0
  13. neuronxcc-2.12.68.0+4480452af/MODULE_ec82f4eab0e71478ee31+2c2d707e/compile_flags.txt +1 -0
  14. neuronxcc-2.12.68.0+4480452af/MODULE_ec82f4eab0e71478ee31+2c2d707e/model.hlo.pb +3 -0
  15. neuronxcc-2.12.68.0+4480452af/MODULE_ec82f4eab0e71478ee31+2c2d707e/model.neff +3 -0
.gitattributes CHANGED
@@ -1639,3 +1639,4 @@ neuronxcc-2.12.68.0+4480452af/MODULE_13c79645c9d868bb2806+2c2d707e/model.neff fi
1639
  neuronxcc-2.12.68.0+4480452af/MODULE_264681f3d540f16615a1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1640
  neuronxcc-2.12.68.0+4480452af/MODULE_04e52f3ca303bb237f59+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1641
  neuronxcc-2.12.68.0+4480452af/MODULE_623dbb4dc9e304b53c18+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 
 
1639
  neuronxcc-2.12.68.0+4480452af/MODULE_264681f3d540f16615a1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1640
  neuronxcc-2.12.68.0+4480452af/MODULE_04e52f3ca303bb237f59+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1641
  neuronxcc-2.12.68.0+4480452af/MODULE_623dbb4dc9e304b53c18+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1642
+ neuronxcc-2.12.68.0+4480452af/MODULE_ec82f4eab0e71478ee31+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.20/inference/llama/01-ai/Yi-9B/95ef5301db7f1b728496.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 64000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 48, "num_attention_heads": 32, "num_key_value_heads": 4, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-06, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "pad_token_id": 0, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "01-ai/Yi-9B", "checkpoint_revision": "58cdd0b4d4cdd2189bbf29a5773c73926084b753", "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/gpt2/gpt2/3bcf76c897d15e9f8e43.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}
neuronxcc-2.12.68.0+4480452af/MODULE_a4b5e92c38528bfc2597+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.12.68.0+4480452af/MODULE_a4b5e92c38528bfc2597+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b6b5e48ce7dce608b5e50446a8e8b2ce7c75ffc957b7bc6a043d48910e177de4
3
+ size 132432
neuronxcc-2.12.68.0+4480452af/MODULE_a4b5e92c38528bfc2597+2c2d707e/model.neff ADDED
Binary file (923 kB). View file
 
neuronxcc-2.12.68.0+4480452af/MODULE_c92166730ec03623bbc8+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.12.68.0+4480452af/MODULE_c92166730ec03623bbc8+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0142ae15f2f3f55c9af19cf779d504e863779e3f71d41d05e90008977ce0a5d4
3
+ size 132432
neuronxcc-2.12.68.0+4480452af/MODULE_c92166730ec03623bbc8+2c2d707e/model.neff ADDED
Binary file (626 kB). View file
 
neuronxcc-2.12.68.0+4480452af/MODULE_ce6900904f4375e48d21+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.12.68.0+4480452af/MODULE_ce6900904f4375e48d21+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cf86da3773ed9f8e6006434f500a2e9d03e7f14e9c0d1c83d4b2aaf737c3ad91
3
+ size 132432
neuronxcc-2.12.68.0+4480452af/MODULE_ce6900904f4375e48d21+2c2d707e/model.neff ADDED
Binary file (728 kB). View file
 
neuronxcc-2.12.68.0+4480452af/MODULE_ec82f4eab0e71478ee31+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.12.68.0+4480452af/MODULE_ec82f4eab0e71478ee31+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8bddd0b3cc10a04ff7dd52c2c6240e271e9b0822c7dd88b2ae09c99e91c2af1
3
+ size 132432
neuronxcc-2.12.68.0+4480452af/MODULE_ec82f4eab0e71478ee31+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2787292829a8e122ffade75d06e5222b4b5fc22e7227af2bbfeeaa465f99be58
3
+ size 1301504