Merge branch 'master' into compilade/mamba2
This commit is contained in:
Francis Couture-Harpin
commit
0e601cafe9
197 changed files with 25587 additions and 16246 deletions
|
|
@ -13,7 +13,7 @@ class TensorNameMap:
|
|||
"transformer.wte", # gpt2 gpt-j mpt refact qwen dbrx jais exaone
|
||||
"transformer.word_embeddings", # falcon
|
||||
"word_embeddings", # bloom
|
||||
"model.embed_tokens", # llama-hf nemotron
|
||||
"model.embed_tokens", # llama-hf nemotron olmoe
|
||||
"tok_embeddings", # llama-pth
|
||||
"embeddings.word_embeddings", # bert nomic-bert
|
||||
"language_model.embedding.word_embeddings", # persimmon
|
||||
|
|
@ -27,6 +27,7 @@ class TensorNameMap:
|
|||
"embedding.word_embeddings", # chatglm
|
||||
"transformer.token_embeddings", # openelm
|
||||
"shared", # t5
|
||||
"rwkv.embeddings", # rwkv
|
||||
),
|
||||
|
||||
# Token type embeddings
|
||||
|
|
@ -40,6 +41,7 @@ class TensorNameMap:
|
|||
"embeddings.LayerNorm", # bert
|
||||
"emb_ln", # nomic-bert
|
||||
"transformer.norm", # openelm
|
||||
"rwkv.blocks.0.pre_ln", # rwkv
|
||||
),
|
||||
|
||||
# Position embeddings
|
||||
|
|
@ -52,18 +54,19 @@ class TensorNameMap:
|
|||
# Output
|
||||
MODEL_TENSOR.OUTPUT: (
|
||||
"embed_out", # gptneox
|
||||
"lm_head", # gpt2 mpt falcon llama-hf baichuan qwen mamba dbrx jais nemotron exaone
|
||||
"lm_head", # gpt2 mpt falcon llama-hf baichuan qwen mamba dbrx jais nemotron exaone olmoe
|
||||
"output", # llama-pth bloom internlm2
|
||||
"word_embeddings_for_head", # persimmon
|
||||
"lm_head.linear", # phi2
|
||||
"output_layer", # chatglm
|
||||
"head", # rwkv
|
||||
),
|
||||
|
||||
# Output norm
|
||||
MODEL_TENSOR.OUTPUT_NORM: (
|
||||
"gpt_neox.final_layer_norm", # gptneox
|
||||
"transformer.ln_f", # gpt2 gpt-j falcon jais exaone
|
||||
"model.norm", # llama-hf baichuan internlm2
|
||||
"model.norm", # llama-hf baichuan internlm2 olmoe
|
||||
"norm", # llama-pth
|
||||
"transformer.norm_f", # mpt dbrx
|
||||
"ln_f", # refact bloom qwen gpt2
|
||||
|
|
@ -76,6 +79,7 @@ class TensorNameMap:
|
|||
"encoder.final_layernorm", # chatglm
|
||||
"transformer.norm", # openelm
|
||||
"model.norm", # nemotron
|
||||
"rwkv.ln_out", # rwkv
|
||||
),
|
||||
|
||||
# Rope frequencies
|
||||
|
|
@ -94,7 +98,7 @@ class TensorNameMap:
|
|||
"transformer.h.{bid}.input_layernorm", # falcon7b
|
||||
"h.{bid}.input_layernorm", # bloom
|
||||
"transformer.h.{bid}.ln_mlp", # falcon40b
|
||||
"model.layers.{bid}.input_layernorm", # llama-hf nemotron
|
||||
"model.layers.{bid}.input_layernorm", # llama-hf nemotron olmoe
|
||||
"layers.{bid}.attention_norm", # llama-pth
|
||||
"language_model.encoder.layers.{bid}.input_layernorm", # persimmon
|
||||
"model.layers.{bid}.ln1", # yi
|
||||
|
|
@ -108,12 +112,14 @@ class TensorNameMap:
|
|||
"transformer.blocks.{bid}.norm_attn_norm.norm_1", # dbrx
|
||||
"encoder.layers.{bid}.input_layernorm", # chatglm
|
||||
"transformer.layers.{bid}.attn_norm", # openelm
|
||||
"rwkv.blocks.{bid}.ln1", # rwkv
|
||||
),
|
||||
|
||||
# Attention norm 2
|
||||
MODEL_TENSOR.ATTN_NORM_2: (
|
||||
"transformer.h.{bid}.ln_attn", # falcon40b
|
||||
"transformer.h.{bid}.ln_attn", # falcon40b
|
||||
"encoder.layer.{bid}.layer_norm_1", # jina-v2-code
|
||||
"rwkv.blocks.{bid}.ln2", # rwkv
|
||||
),
|
||||
|
||||
# Attention query-key-value
|
||||
|
|
@ -136,7 +142,7 @@ class TensorNameMap:
|
|||
|
||||
# Attention query
|
||||
MODEL_TENSOR.ATTN_Q: (
|
||||
"model.layers.{bid}.self_attn.q_proj", # llama-hf nemotron
|
||||
"model.layers.{bid}.self_attn.q_proj", # llama-hf nemotron olmoe
|
||||
"layers.{bid}.attention.wq", # llama-pth
|
||||
"encoder.layer.{bid}.attention.self.query", # bert
|
||||
"transformer.h.{bid}.attn.q_proj", # gpt-j
|
||||
|
|
@ -148,7 +154,7 @@ class TensorNameMap:
|
|||
|
||||
# Attention key
|
||||
MODEL_TENSOR.ATTN_K: (
|
||||
"model.layers.{bid}.self_attn.k_proj", # llama-hf nemotron
|
||||
"model.layers.{bid}.self_attn.k_proj", # llama-hf nemotron olmoe
|
||||
"layers.{bid}.attention.wk", # llama-pth
|
||||
"encoder.layer.{bid}.attention.self.key", # bert
|
||||
"transformer.h.{bid}.attn.k_proj", # gpt-j
|
||||
|
|
@ -161,7 +167,7 @@ class TensorNameMap:
|
|||
|
||||
# Attention value
|
||||
MODEL_TENSOR.ATTN_V: (
|
||||
"model.layers.{bid}.self_attn.v_proj", # llama-hf nemotron
|
||||
"model.layers.{bid}.self_attn.v_proj", # llama-hf nemotron olmoe
|
||||
"layers.{bid}.attention.wv", # llama-pth
|
||||
"encoder.layer.{bid}.attention.self.value", # bert
|
||||
"transformer.h.{bid}.attn.v_proj", # gpt-j
|
||||
|
|
@ -179,7 +185,7 @@ class TensorNameMap:
|
|||
"transformer.blocks.{bid}.attn.out_proj", # mpt
|
||||
"transformer.h.{bid}.self_attention.dense", # falcon
|
||||
"h.{bid}.self_attention.dense", # bloom
|
||||
"model.layers.{bid}.self_attn.o_proj", # llama-hf nemotron
|
||||
"model.layers.{bid}.self_attn.o_proj", # llama-hf nemotron olmoe
|
||||
"layers.{bid}.attention.wo", # llama-pth
|
||||
"encoder.layer.{bid}.attention.output.dense", # bert
|
||||
"transformer.h.{bid}.attn.out_proj", # gpt-j
|
||||
|
|
@ -223,7 +229,7 @@ class TensorNameMap:
|
|||
"transformer.h.{bid}.ln_2", # gpt2 refact qwen jais exaone
|
||||
"h.{bid}.post_attention_layernorm", # bloom
|
||||
"transformer.blocks.{bid}.norm_2", # mpt
|
||||
"model.layers.{bid}.post_attention_layernorm", # llama-hf nemotron
|
||||
"model.layers.{bid}.post_attention_layernorm", # llama-hf nemotron olmoe
|
||||
"layers.{bid}.ffn_norm", # llama-pth
|
||||
"language_model.encoder.layers.{bid}.post_attention_layernorm", # persimmon
|
||||
"model.layers.{bid}.ln2", # yi
|
||||
|
|
@ -247,7 +253,7 @@ class TensorNameMap:
|
|||
MODEL_TENSOR.FFN_GATE_INP: (
|
||||
"layers.{bid}.feed_forward.gate", # mixtral
|
||||
"model.layers.{bid}.block_sparse_moe.gate", # mixtral
|
||||
"model.layers.{bid}.mlp.gate", # qwen2moe
|
||||
"model.layers.{bid}.mlp.gate", # qwen2moe olmoe
|
||||
"transformer.decoder_layer.{bid}.router", # Grok
|
||||
"transformer.blocks.{bid}.ffn.router.layer", # dbrx
|
||||
),
|
||||
|
|
@ -289,7 +295,7 @@ class TensorNameMap:
|
|||
"layers.{bid}.feed_forward.experts.w3", # mixtral (merged)
|
||||
"transformer.decoder_layer.{bid}.moe.linear_v", # Grok (merged)
|
||||
"transformer.blocks.{bid}.ffn.experts.mlp.v1", # dbrx
|
||||
"model.layers.{bid}.mlp.experts.up_proj", # qwen2moe (merged)
|
||||
"model.layers.{bid}.mlp.experts.up_proj", # qwen2moe olmoe (merged)
|
||||
),
|
||||
|
||||
MODEL_TENSOR.FFN_UP_SHEXP: (
|
||||
|
|
@ -321,7 +327,7 @@ class TensorNameMap:
|
|||
"layers.{bid}.feed_forward.experts.w1", # mixtral (merged)
|
||||
"transformer.decoder_layer.{bid}.moe.linear", # Grok (merged)
|
||||
"transformer.blocks.{bid}.ffn.experts.mlp.w1", # dbrx
|
||||
"model.layers.{bid}.mlp.experts.gate_proj", # qwen2moe (merged)
|
||||
"model.layers.{bid}.mlp.experts.gate_proj", # qwen2moe olmoe (merged)
|
||||
),
|
||||
|
||||
MODEL_TENSOR.FFN_GATE_SHEXP: (
|
||||
|
|
@ -361,7 +367,7 @@ class TensorNameMap:
|
|||
"layers.{bid}.feed_forward.experts.w2", # mixtral (merged)
|
||||
"transformer.decoder_layer.{bid}.moe.linear_1", # Grok (merged)
|
||||
"transformer.blocks.{bid}.ffn.experts.mlp.w2", # dbrx
|
||||
"model.layers.{bid}.mlp.experts.down_proj", # qwen2moe (merged)
|
||||
"model.layers.{bid}.mlp.experts.down_proj", # qwen2moe olmoe (merged)
|
||||
),
|
||||
|
||||
MODEL_TENSOR.FFN_DOWN_SHEXP: (
|
||||
|
|
@ -372,7 +378,7 @@ class TensorNameMap:
|
|||
MODEL_TENSOR.ATTN_Q_NORM: (
|
||||
"language_model.encoder.layers.{bid}.self_attention.q_layernorm",
|
||||
"model.layers.{bid}.self_attn.q_layernorm", # persimmon
|
||||
"model.layers.{bid}.self_attn.q_norm", # cohere
|
||||
"model.layers.{bid}.self_attn.q_norm", # cohere olmoe
|
||||
"transformer.blocks.{bid}.attn.q_ln", # sea-lion
|
||||
"encoder.layer.{bid}.attention.self.layer_norm_q", # jina-bert-v2
|
||||
"transformer.layers.{bid}.attn.q_norm", # openelm
|
||||
|
|
@ -381,7 +387,7 @@ class TensorNameMap:
|
|||
MODEL_TENSOR.ATTN_K_NORM: (
|
||||
"language_model.encoder.layers.{bid}.self_attention.k_layernorm",
|
||||
"model.layers.{bid}.self_attn.k_layernorm", # persimmon
|
||||
"model.layers.{bid}.self_attn.k_norm", # cohere
|
||||
"model.layers.{bid}.self_attn.k_norm", # cohere olmoe
|
||||
"transformer.blocks.{bid}.attn.k_ln", # sea-lion
|
||||
"encoder.layer.{bid}.attention.self.layer_norm_k", # jina-bert-v2
|
||||
"transformer.layers.{bid}.attn.k_norm", # openelm
|
||||
|
|
@ -438,6 +444,98 @@ class TensorNameMap:
|
|||
"backbone.layers.{bid}.mixer.out_proj",
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_W1: (
|
||||
"rwkv.blocks.{bid}.attention.time_maa_w1", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_W2: (
|
||||
"rwkv.blocks.{bid}.attention.time_maa_w2", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_LERP_X: (
|
||||
"rwkv.blocks.{bid}.attention.time_maa_x", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_LERP_K: (
|
||||
"rwkv.blocks.{bid}.attention.time_maa_k", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_LERP_V: (
|
||||
"rwkv.blocks.{bid}.attention.time_maa_v", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_LERP_R: (
|
||||
"rwkv.blocks.{bid}.attention.time_maa_r", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_LERP_G: (
|
||||
"rwkv.blocks.{bid}.attention.time_maa_g", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_LERP_W: (
|
||||
"rwkv.blocks.{bid}.attention.time_maa_w", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_FIRST: (
|
||||
"rwkv.blocks.{bid}.attention.time_faaaa", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_DECAY: (
|
||||
"rwkv.blocks.{bid}.attention.time_decay", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_DECAY_W1: (
|
||||
"rwkv.blocks.{bid}.attention.time_decay_w1", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_DECAY_W2: (
|
||||
"rwkv.blocks.{bid}.attention.time_decay_w2", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_KEY: (
|
||||
"rwkv.blocks.{bid}.attention.key", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_VALUE: (
|
||||
"rwkv.blocks.{bid}.attention.value", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_RECEPTANCE: (
|
||||
"rwkv.blocks.{bid}.attention.receptance", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_GATE: (
|
||||
"rwkv.blocks.{bid}.attention.gate", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_LN: (
|
||||
"rwkv.blocks.{bid}.attention.ln_x", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.TIME_MIX_OUTPUT: (
|
||||
"rwkv.blocks.{bid}.attention.output", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.CHANNEL_MIX_LERP_K: (
|
||||
"rwkv.blocks.{bid}.feed_forward.time_maa_k", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.CHANNEL_MIX_LERP_R: (
|
||||
"rwkv.blocks.{bid}.feed_forward.time_maa_r", # rwkv v6
|
||||
),
|
||||
|
||||
MODEL_TENSOR.CHANNEL_MIX_KEY: (
|
||||
"rwkv.blocks.{bid}.feed_forward.key", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.CHANNEL_MIX_RECEPTANCE: (
|
||||
"rwkv.blocks.{bid}.feed_forward.receptance", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.CHANNEL_MIX_VALUE: (
|
||||
"rwkv.blocks.{bid}.feed_forward.value", # rwkv
|
||||
),
|
||||
|
||||
MODEL_TENSOR.ATTN_Q_A: (
|
||||
"model.layers.{bid}.self_attn.q_a_proj", # deepseek2
|
||||
),
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue