From 476b0251b27fb64c575507024a671e639d675594 Mon Sep 17 00:00:00 2001 From: Julius Arkenberg Date: Sat, 23 Mar 2024 17:41:53 +0100 Subject: llama : add grok-1 support (#6204) * Add support for Grok model architecture * Revert convert-hf-to-gguf to default options * Fixed f_norm_rms_eps bug * Fix whitespaces * llama : fix grok rope type * llama : minor --------- Co-authored-by: Georgi Gerganov --- gguf-py/gguf/constants.py | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) (limited to 'gguf-py/gguf/constants.py') diff --git a/gguf-py/gguf/constants.py b/gguf-py/gguf/constants.py index 4a4facb0..e47896e2 100644 --- a/gguf-py/gguf/constants.py +++ b/gguf-py/gguf/constants.py @@ -100,6 +100,7 @@ class MODEL_ARCH(IntEnum): LLAMA = auto() FALCON = auto() BAICHUAN = auto() + GROK = auto() GPT2 = auto() GPTJ = auto() GPTNEOX = auto() @@ -167,6 +168,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = { MODEL_ARCH.LLAMA: "llama", MODEL_ARCH.FALCON: "falcon", MODEL_ARCH.BAICHUAN: "baichuan", + MODEL_ARCH.GROK: "grok", MODEL_ARCH.GPT2: "gpt2", MODEL_ARCH.GPTJ: "gptj", MODEL_ARCH.GPTNEOX: "gptneox", @@ -251,6 +253,28 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = { MODEL_TENSOR.FFN_DOWN_EXP, MODEL_TENSOR.FFN_UP_EXP, ], + MODEL_ARCH.GROK: [ + MODEL_TENSOR.TOKEN_EMBD, + MODEL_TENSOR.OUTPUT_NORM, + MODEL_TENSOR.OUTPUT, + MODEL_TENSOR.ROPE_FREQS, + MODEL_TENSOR.ATTN_NORM, + MODEL_TENSOR.ATTN_Q, + MODEL_TENSOR.ATTN_K, + MODEL_TENSOR.ATTN_V, + MODEL_TENSOR.ATTN_OUT, + MODEL_TENSOR.ATTN_ROT_EMBD, + MODEL_TENSOR.ATTN_OUT_NORM, + MODEL_TENSOR.FFN_GATE_INP, + MODEL_TENSOR.FFN_NORM, + MODEL_TENSOR.FFN_GATE, + MODEL_TENSOR.FFN_DOWN, + MODEL_TENSOR.FFN_UP, + MODEL_TENSOR.FFN_GATE_EXP, + MODEL_TENSOR.FFN_DOWN_EXP, + MODEL_TENSOR.FFN_UP_EXP, + MODEL_TENSOR.LAYER_OUT_NORM, + ], MODEL_ARCH.GPTNEOX: [ MODEL_TENSOR.TOKEN_EMBD, MODEL_TENSOR.OUTPUT_NORM, -- cgit v1.2.3