From 2b3b999cacc7ad1207c32fbdf3479a19c06e1a34 Mon Sep 17 00:00:00 2001 From: chiranko <96988916+chiranko@users.noreply.github.com> Date: Fri, 19 Jan 2024 17:07:27 +0800 Subject: llama : add CodeShell support (#5016) * llama: add codeshell support * llama.cpp: fix codeshell with NeoX rope Co-authored-by: Georgi Gerganov --------- Co-authored-by: Georgi Gerganov --- gguf-py/gguf/constants.py | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) (limited to 'gguf-py/gguf/constants.py') 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, + ], } # -- cgit v1.2.3