root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-29 04:44:34 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

convert-hf : support Mini-Jamba conversion

Browse Source
This commit is contained in:
Francis Couture-Harpin 2024-05-25 13:55:11 -04:00
parent ea2e63e9d2
commit fc59407efe
2 changed files with 23 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
convert-hf-to-gguf.py
View File

@ -2393,6 +2393,16 @@ class JambaModel(Model):
return "gpt-2" return "gpt-2"
def set_vocab(self):
if (self.dir_model / "tokenizer.model").is_file():
# Using Jamba's tokenizer.json causes errors on model load
# (something about "byte not found in vocab"),
# but there's a working tokenizer.model
self._set_vocab_sentencepiece()
else:
# Some Jamba models only have a tokenizer.json, which works.
self._set_vocab_gpt2()
def set_gguf_parameters(self): def set_gguf_parameters(self):
d_model = self.find_hparam(["hidden_size", "mamba_d_model"]) d_model = self.find_hparam(["hidden_size", "mamba_d_model"])
d_conv = self.find_hparam(["mamba_d_conv"], optional=True) or 4 d_conv = self.find_hparam(["mamba_d_conv"], optional=True) or 4
@ -2412,7 +2422,7 @@ class JambaModel(Model):
self.gguf_writer.add_name(self.dir_model.name) self.gguf_writer.add_name(self.dir_model.name)
self.gguf_writer.add_block_count(self.block_count) self.gguf_writer.add_block_count(self.block_count)
self.gguf_writer.add_context_length(self.hparams["max_position_embeddings"]) self.gguf_writer.add_context_length(self.find_hparam(["max_position_embeddings", "n_ctx"]))
self.gguf_writer.add_embedding_length(d_model) self.gguf_writer.add_embedding_length(d_model)
self.gguf_writer.add_feed_forward_length(self.hparams["intermediate_size"]) self.gguf_writer.add_feed_forward_length(self.hparams["intermediate_size"])
self.gguf_writer.add_head_count(self.hparams["num_attention_heads"]) self.gguf_writer.add_head_count(self.hparams["num_attention_heads"])
@ -2430,6 +2440,15 @@ class JambaModel(Model):
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]: def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
# Mini-Jamba
name = name.replace(".moe.", ".feed_forward.")
if bid is not None:
moe_offset = self.hparams["expert_layer_offset"]
moe_period = self.hparams["expert_layer_period"]
if not (bid >= moe_offset and (bid - moe_offset) % moe_period == 0):
name = name.replace(".experts.0.", ".")
# process the experts separately # process the experts separately
if ".feed_forward.experts." in name: if ".feed_forward.experts." in name:
n_experts = self.hparams["num_experts"] n_experts = self.hparams["num_experts"]

3
gguf-py/gguf/tensor_mapping.py
View File

@ -207,6 +207,7 @@ class TensorNameMap:
"model.layers.{bid}.ffn_norm", # internlm2 "model.layers.{bid}.ffn_norm", # internlm2
"transformer.decoder_layer.{bid}.rms_norm_2", # Grok "transformer.decoder_layer.{bid}.rms_norm_2", # Grok
"model.layers.{bid}.pre_ff_layernorm", # jamba "model.layers.{bid}.pre_ff_layernorm", # jamba
"model.layers.{bid}.pre_moe_layernorm", # mini-jamba
), ),
MODEL_TENSOR.FFN_GATE_INP: ( MODEL_TENSOR.FFN_GATE_INP: (
@ -390,10 +391,12 @@ class TensorNameMap:
MODEL_TENSOR.SSM_B_NORM: ( MODEL_TENSOR.SSM_B_NORM: (
"model.layers.{bid}.mamba.b_layernorm", # jamba "model.layers.{bid}.mamba.b_layernorm", # jamba
"model.layers.{bid}.mamba.B_layernorm", # mini-jamba
), ),
MODEL_TENSOR.SSM_C_NORM: ( MODEL_TENSOR.SSM_C_NORM: (
"model.layers.{bid}.mamba.c_layernorm", # jamba "model.layers.{bid}.mamba.c_layernorm", # jamba
"model.layers.{bid}.mamba.C_layernorm", # mini-jamba
), ),
MODEL_TENSOR.SSM_D: ( MODEL_TENSOR.SSM_D: (