convert : more consistent handling of rope_parameters (#24833)
This commit is contained in:
+9
-11
@@ -18,7 +18,7 @@ class Phi2Model(TextModel):
|
||||
model_arch = gguf.MODEL_ARCH.PHI2
|
||||
|
||||
def set_gguf_parameters(self):
|
||||
rot_pct = self.find_hparam(["partial_rotary_factor"])
|
||||
rot_pct = self.rope_parameters["partial_rotary_factor"]
|
||||
n_embd = self.find_hparam(["hidden_size", "n_embd"])
|
||||
n_head = self.find_hparam(["num_attention_heads", "n_head"])
|
||||
|
||||
@@ -149,8 +149,8 @@ class Phi3MiniModel(TextModel):
|
||||
n_head_kv = self.find_hparam(["num_key_value_heads", "n_head_kv"])
|
||||
rms_eps = self.find_hparam(["rms_norm_eps"])
|
||||
max_pos_embds = self.find_hparam(["n_positions", "max_position_embeddings"])
|
||||
orig_max_pos_embds = self.find_hparam(["original_max_position_embeddings"])
|
||||
rot_pct = self.hparams.get("partial_rotary_factor", 1.0)
|
||||
orig_max_pos_embds = self.rope_parameters["original_max_position_embeddings"]
|
||||
rot_pct = self.rope_parameters.get("partial_rotary_factor", 1.0)
|
||||
rope_dims = int(rot_pct * n_embd) // n_head
|
||||
|
||||
self.gguf_writer.add_context_length(max_pos_embds)
|
||||
@@ -174,18 +174,19 @@ class Phi3MiniModel(TextModel):
|
||||
n_embd = self.find_hparam(["hidden_size", "n_embd"])
|
||||
n_head = self.find_hparam(["num_attention_heads", "n_head"])
|
||||
max_pos_embds = self.find_hparam(["n_positions", "max_position_embeddings"])
|
||||
orig_max_pos_embds = self.find_hparam(["original_max_position_embeddings"])
|
||||
rot_pct = self.hparams.get("partial_rotary_factor", 1.0)
|
||||
orig_max_pos_embds = self.rope_parameters["original_max_position_embeddings"]
|
||||
rot_pct = self.rope_parameters.get("partial_rotary_factor", 1.0)
|
||||
rope_dims = int(rot_pct * n_embd) // n_head
|
||||
|
||||
# write rope scaling for long context (128k) model
|
||||
rope_scaling = self.find_hparam(['rope_scaling'], True)
|
||||
if rope_scaling is None:
|
||||
long_factors = self.rope_parameters.get('long_factor')
|
||||
short_factors = self.rope_parameters.get('short_factor')
|
||||
if not long_factors:
|
||||
return
|
||||
|
||||
scale = max_pos_embds / orig_max_pos_embds
|
||||
|
||||
rope_scaling_type = rope_scaling.get('rope_type', rope_scaling.get('type', '')).lower()
|
||||
rope_scaling_type = self.rope_parameters.get('rope_type', '').lower()
|
||||
if len(rope_scaling_type) == 0:
|
||||
raise KeyError('Missing the required key rope_scaling.type')
|
||||
|
||||
@@ -198,9 +199,6 @@ class Phi3MiniModel(TextModel):
|
||||
|
||||
self.gguf_writer.add_rope_scaling_attn_factors(attn_factor)
|
||||
|
||||
long_factors = rope_scaling.get('long_factor', None)
|
||||
short_factors = rope_scaling.get('short_factor', None)
|
||||
|
||||
if long_factors is None or short_factors is None:
|
||||
raise KeyError('Missing the required key rope_scaling.long_factor or rope_scaling_short_factor')
|
||||
|
||||
|
||||
Reference in New Issue
Block a user