vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

gguf : add rope_freq_base parameter for CodeLlama (#2769)

Browse source
This commit is contained in:
slaren 2023-08-24 20:04:05 +02:00 • committed by GitHub
parent 01f2224682
commit 0d3094f0c7
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 48 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

43
convert.py
View file

@ -104,6 +104,8 @@ class Params:
n_head_kv: int
f_norm_eps: float
f_rope_freq_base: Optional[float] = None
ftype: Optional[GGMLFileType] = None
# path to the directory containing the model files
@ -194,15 +196,16 @@ class Params:
def loadOriginalParamsJson(model: 'LazyModel', config_path: 'Path') -> 'Params':
config = json.load(open(config_path))
n_vocab = config["vocab_size"] if "vocab_size" in config else -1
n_embd = config["dim"]
n_layer = config["n_layers"]
n_mult = config["multiple_of"]
n_ctx = 2048 if config["norm_eps"] == 1e-06 else 4096 # hack to determine LLaMA v1 vs v2
n_ff = -1
n_head = config["n_heads"]
n_head_kv = config["n_kv_heads"] if "n_kv_heads" in config else n_head
f_norm_eps = config["norm_eps"]
n_vocab = config["vocab_size"] if "vocab_size" in config else -1
n_embd = config["dim"]
n_layer = config["n_layers"]
n_mult = config["multiple_of"]
n_ctx = 2048 if config["norm_eps"] == 1e-06 else 4096 # hack to determine LLaMA v1 vs v2
n_ff = -1
n_head = config["n_heads"]
n_head_kv = config["n_kv_heads"] if "n_kv_heads" in config else n_head
f_norm_eps = config["norm_eps"]
f_rope_freq_base = config["rope_theta"] if "rope_theta" in config else None
if n_vocab == -1:
n_vocab = model["tok_embeddings.weight"].shape[0]
@ -211,15 +214,16 @@ class Params:
n_ff = model["layers.0.feed_forward.w1.weight"].shape[0]
return Params(
n_vocab = n_vocab,
n_embd = n_embd,
n_mult = n_mult,
n_layer = n_layer,
n_ctx = n_ctx,
n_ff = n_ff,
n_head = n_head,
n_head_kv = n_head_kv,
f_norm_eps = f_norm_eps,
n_vocab = n_vocab,
n_embd = n_embd,
n_mult = n_mult,
n_layer = n_layer,
n_ctx = n_ctx,
n_ff = n_ff,
n_head = n_head,
n_head_kv = n_head_kv,
f_norm_eps = f_norm_eps,
f_rope_freq_base = f_rope_freq_base,
)
@staticmethod
@ -754,6 +758,9 @@ class OutputFile:
self.gguf.add_head_count_kv (params.n_head_kv)
self.gguf.add_layer_norm_rms_eps (params.f_norm_eps)
if params.f_rope_freq_base:
self.gguf.add_rope_freq_base(params.f_rope_freq_base)
if params.ftype:
self.gguf.add_file_type(params.ftype)