| { |
| "version": "stemma-index-v1", |
| "dim": 1456, |
| "metric": "cosine", |
| "backend": "faiss", |
| "ids": [ |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-135m-root", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/gpt2-root", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/distilgpt2", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-sft", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-lora-merged", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-cpt", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-int8", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-sft-int4", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/gpt2-int8", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-prune-mag30", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-sft-prune-struct10", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-vocab-ext", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-vocab-ext-trained", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-merge-slerp", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-merge-ties2", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-merge-ties3", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-merge-dare2", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-distil-half", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/control-rand-init-a", |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/control-rand-init-b" |
| ], |
| "meta": { |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-135m-root": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/gpt2-root": { |
| "n_layers": 12, |
| "hidden_size": 768, |
| "vocab_size": 50257, |
| "dtypes": { |
| "F32": 160 |
| }, |
| "architectures": [ |
| "GPT2LMHeadModel" |
| ], |
| "tensor_count": 160, |
| "param_count": 137022720, |
| "fused_qkv": true, |
| "intermediate_size": 3072, |
| "n_1d_tensors": 98, |
| "q_shape": [ |
| 768, |
| 2304 |
| ], |
| "kv_shape": [], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_o@0": "h.0.attn.c_proj.weight", |
| "attn_q@0": "h.0.attn.c_attn.weight", |
| "embed@0": "wte.weight", |
| "mlp_in@0": "h.0.mlp.c_fc.weight", |
| "mlp_out@0": "h.0.mlp.c_proj.weight", |
| "norm@4": "h.0.ln_1.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/distilgpt2": { |
| "n_layers": 6, |
| "hidden_size": 768, |
| "vocab_size": 50257, |
| "dtypes": { |
| "F32": 82 |
| }, |
| "architectures": [ |
| "GPT2LMHeadModel" |
| ], |
| "tensor_count": 82, |
| "param_count": 88204032, |
| "fused_qkv": true, |
| "intermediate_size": 3072, |
| "n_1d_tensors": 50, |
| "q_shape": [ |
| 768, |
| 2304 |
| ], |
| "kv_shape": [], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_o@0": "transformer.h.0.attn.c_proj.weight", |
| "attn_o@1": "transformer.h.1.attn.c_proj.weight", |
| "attn_o@2": "transformer.h.2.attn.c_proj.weight", |
| "attn_o@3": "transformer.h.4.attn.c_proj.weight", |
| "attn_o@4": "transformer.h.5.attn.c_proj.weight", |
| "attn_q@0": "transformer.h.0.attn.c_attn.weight", |
| "attn_q@1": "transformer.h.1.attn.c_attn.weight", |
| "attn_q@2": "transformer.h.2.attn.c_attn.weight", |
| "attn_q@3": "transformer.h.4.attn.c_attn.weight", |
| "attn_q@4": "transformer.h.5.attn.c_attn.weight", |
| "embed@0": "transformer.wte.weight", |
| "mlp_in@0": "transformer.h.0.mlp.c_fc.weight", |
| "mlp_in@1": "transformer.h.1.mlp.c_fc.weight", |
| "mlp_in@2": "transformer.h.2.mlp.c_fc.weight", |
| "mlp_in@3": "transformer.h.4.mlp.c_fc.weight", |
| "mlp_in@4": "transformer.h.5.mlp.c_fc.weight", |
| "mlp_out@0": "transformer.h.0.mlp.c_proj.weight", |
| "mlp_out@1": "transformer.h.1.mlp.c_proj.weight", |
| "mlp_out@2": "transformer.h.2.mlp.c_proj.weight", |
| "mlp_out@3": "transformer.h.4.mlp.c_proj.weight", |
| "mlp_out@4": "transformer.h.5.mlp.c_proj.weight", |
| "norm@0": "transformer.h.0.ln_1.weight", |
| "norm@1": "transformer.h.1.ln_1.weight", |
| "norm@2": "transformer.h.2.ln_1.weight", |
| "norm@3": "transformer.h.4.ln_1.weight", |
| "norm@4": "transformer.h.5.ln_1.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-sft": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-lora-merged": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-cpt": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-int8": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "F16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-sft-int4": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "F16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/gpt2-int8": { |
| "n_layers": 12, |
| "hidden_size": 768, |
| "vocab_size": 50257, |
| "dtypes": { |
| "F16": 160 |
| }, |
| "architectures": [ |
| "GPT2LMHeadModel" |
| ], |
| "tensor_count": 160, |
| "param_count": 137022720, |
| "fused_qkv": true, |
| "intermediate_size": 3072, |
| "n_1d_tensors": 98, |
| "q_shape": [ |
| 768, |
| 2304 |
| ], |
| "kv_shape": [], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_o@0": "h.0.attn.c_proj.weight", |
| "attn_q@0": "h.0.attn.c_attn.weight", |
| "embed@0": "wte.weight", |
| "mlp_in@0": "h.0.mlp.c_fc.weight", |
| "mlp_out@0": "h.0.mlp.c_proj.weight", |
| "norm@4": "h.0.ln_1.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-prune-mag30": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-sft-prune-struct10": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-vocab-ext": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49664, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134809920, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-vocab-ext-trained": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49664, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134809920, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-merge-slerp": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-merge-ties2": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-merge-ties3": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-merge-dare2": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/smollm2-distil-half": { |
| "n_layers": 15, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 137 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 137, |
| "param_count": 81413568, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 31, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.2.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.6.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.13.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.2.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.6.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.13.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.2.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.6.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.13.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.2.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.6.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.13.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.2.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.6.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.13.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.2.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.6.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.13.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.2.input_layernorm.weight", |
| "norm@2": "model.layers.6.input_layernorm.weight", |
| "norm@3": "model.layers.10.input_layernorm.weight", |
| "norm@4": "model.layers.13.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/control-rand-init-a": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| }, |
| "/Users/nagaoyuta/Desktop/Claude code/24-Stemma/bench_models/control-rand-init-b": { |
| "n_layers": 30, |
| "hidden_size": 576, |
| "vocab_size": 49152, |
| "dtypes": { |
| "BF16": 272 |
| }, |
| "architectures": [ |
| "LlamaForCausalLM" |
| ], |
| "tensor_count": 272, |
| "param_count": 134515008, |
| "fused_qkv": false, |
| "intermediate_size": 1536, |
| "n_1d_tensors": 61, |
| "q_shape": [ |
| 576, |
| 576 |
| ], |
| "kv_shape": [ |
| 192, |
| 576 |
| ], |
| "tied_embeddings": true, |
| "slot_tensors": { |
| "attn_k@0": "model.layers.0.self_attn.k_proj.weight", |
| "attn_k@1": "model.layers.10.self_attn.k_proj.weight", |
| "attn_k@2": "model.layers.11.self_attn.k_proj.weight", |
| "attn_k@3": "model.layers.19.self_attn.k_proj.weight", |
| "attn_k@4": "model.layers.26.self_attn.k_proj.weight", |
| "attn_o@0": "model.layers.0.self_attn.o_proj.weight", |
| "attn_o@1": "model.layers.10.self_attn.o_proj.weight", |
| "attn_o@2": "model.layers.11.self_attn.o_proj.weight", |
| "attn_o@3": "model.layers.19.self_attn.o_proj.weight", |
| "attn_o@4": "model.layers.26.self_attn.o_proj.weight", |
| "attn_q@0": "model.layers.0.self_attn.q_proj.weight", |
| "attn_q@1": "model.layers.10.self_attn.q_proj.weight", |
| "attn_q@2": "model.layers.11.self_attn.q_proj.weight", |
| "attn_q@3": "model.layers.19.self_attn.q_proj.weight", |
| "attn_q@4": "model.layers.26.self_attn.q_proj.weight", |
| "attn_v@0": "model.layers.0.self_attn.v_proj.weight", |
| "attn_v@1": "model.layers.10.self_attn.v_proj.weight", |
| "attn_v@2": "model.layers.11.self_attn.v_proj.weight", |
| "attn_v@3": "model.layers.19.self_attn.v_proj.weight", |
| "attn_v@4": "model.layers.26.self_attn.v_proj.weight", |
| "embed@0": "model.embed_tokens.weight", |
| "mlp_in@0": "model.layers.0.mlp.gate_proj.weight", |
| "mlp_in@1": "model.layers.10.mlp.gate_proj.weight", |
| "mlp_in@2": "model.layers.11.mlp.gate_proj.weight", |
| "mlp_in@3": "model.layers.19.mlp.gate_proj.weight", |
| "mlp_in@4": "model.layers.26.mlp.gate_proj.weight", |
| "mlp_out@0": "model.layers.0.mlp.down_proj.weight", |
| "mlp_out@1": "model.layers.10.mlp.down_proj.weight", |
| "mlp_out@2": "model.layers.11.mlp.down_proj.weight", |
| "mlp_out@3": "model.layers.19.mlp.down_proj.weight", |
| "mlp_out@4": "model.layers.26.mlp.down_proj.weight", |
| "norm@0": "model.layers.0.input_layernorm.weight", |
| "norm@1": "model.layers.10.input_layernorm.weight", |
| "norm@2": "model.layers.11.input_layernorm.weight", |
| "norm@3": "model.layers.19.input_layernorm.weight", |
| "norm@4": "model.layers.26.input_layernorm.weight" |
| }, |
| "max_rows": 2048, |
| "k": 64, |
| "seed": 0 |
| } |
| }, |
| "npz": "sketch_index.npz" |
| } |
