diff options
| author | chiranko <96988916+chiranko@users.noreply.github.com> | 2024-01-19 17:07:27 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-01-19 11:07:27 +0200 |
| commit | 2b3b999cacc7ad1207c32fbdf3479a19c06e1a34 (patch) | |
| tree | be2adbdfb73eb5d24ab9bcbb6f53671b9bb88ffa /gguf-py | |
| parent | 993fba81807e55d27b570945af8e416d535eced1 (diff) | |
llama : add CodeShell support (#5016)
* llama: add codeshell support
* llama.cpp: fix codeshell with NeoX rope
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Diffstat (limited to 'gguf-py')
| -rw-r--r-- | gguf-py/gguf/constants.py | 19 | ||||
| -rw-r--r-- | gguf-py/gguf/tensor_mapping.py | 1 |
2 files changed, 20 insertions, 0 deletions
diff --git a/gguf-py/gguf/constants.py b/gguf-py/gguf/constants.py index 972b4e9a..95c58b41 100644 --- a/gguf-py/gguf/constants.py +++ b/gguf-py/gguf/constants.py @@ -99,6 +99,7 @@ class MODEL_ARCH(IntEnum): QWEN = auto() PHI2 = auto() PLAMO = auto() + CODESHELL = auto() class MODEL_TENSOR(IntEnum): @@ -147,6 +148,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = { MODEL_ARCH.QWEN: "qwen", MODEL_ARCH.PHI2: "phi2", MODEL_ARCH.PLAMO: "plamo", + MODEL_ARCH.CODESHELL: "codeshell", } TENSOR_NAMES: dict[MODEL_TENSOR, str] = { @@ -396,6 +398,19 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = { MODEL_TENSOR.FFN_NORM, MODEL_TENSOR.FFN_DOWN, MODEL_TENSOR.FFN_UP, + ], + MODEL_ARCH.CODESHELL: [ + MODEL_TENSOR.TOKEN_EMBD, + MODEL_TENSOR.POS_EMBD, + MODEL_TENSOR.OUTPUT_NORM, + MODEL_TENSOR.OUTPUT, + MODEL_TENSOR.ATTN_NORM, + MODEL_TENSOR.ATTN_QKV, + MODEL_TENSOR.ATTN_OUT, + MODEL_TENSOR.ATTN_ROT_EMBD, + MODEL_TENSOR.FFN_NORM, + MODEL_TENSOR.FFN_DOWN, + MODEL_TENSOR.FFN_UP, ] # TODO } @@ -417,6 +432,10 @@ MODEL_TENSOR_SKIP: dict[MODEL_ARCH, list[MODEL_TENSOR]] = { MODEL_TENSOR.ROPE_FREQS, MODEL_TENSOR.ATTN_ROT_EMBD, ], + MODEL_ARCH.CODESHELL: [ + MODEL_TENSOR.ROPE_FREQS, + MODEL_TENSOR.ATTN_ROT_EMBD, + ], } # diff --git a/gguf-py/gguf/tensor_mapping.py b/gguf-py/gguf/tensor_mapping.py index e5b14610..de177af1 100644 --- a/gguf-py/gguf/tensor_mapping.py +++ b/gguf-py/gguf/tensor_mapping.py @@ -154,6 +154,7 @@ class TensorNameMap: "model.layers.{bid}.self_attn.rotary_emb.inv_freq", # llama-hf "layers.{bid}.attention.inner_attention.rope.freqs", # llama-pth "model.layers.layers.{bid}.self_attn.rotary_emb.inv_freq", # plamo + "transformer.h.{bid}.attn.rotary_emb.inv_freq", # codeshell ), # Feed-forward norm |