fix: cleanup config, remove unused values and fix non flash init

server/text_generation_server/models/__init__.py

@@ -242,7 +242,13 @@ def get_model(
                 use_medusa=use_medusa,
             )
         else:
-            raise NotImplementedError(FLASH_ATT_ERROR_MESSAGE.format("Phi"))
+            return CausalLM(
+                model_id,
+                revision,
+                quantize=quantize,
+                dtype=dtype,
+                trust_remote_code=trust_remote_code,
+            )
 
     elif model_type == "phi-msft":
         if FLASH_ATTENTION:

server/text_generation_server/models/custom_modeling/flash_phi_modeling.py

@@ -25,34 +25,24 @@ class PhiConfig(PretrainedConfig):
         num_hidden_layers=32,
         num_attention_heads=32,
         num_key_value_heads=32,
-        hidden_act="gelu_fast", # llama uses silu
-        max_position_embeddings=2048,
-        initializer_range=0.02,
-        layer_norm_eps=1e-05, # rms in llama
-        use_cache=True,
+        hidden_act="gelu_fast",    # llama uses silu
+        layer_norm_eps=1e-05,      # rms in llama,
         pad_token_id=0,
         bos_token_id=1,
         eos_token_id=2,
-        pretraining_tp=1,
         tie_word_embeddings=False,
-        rope_scaling=None,
         rope_theta=10000.0,
-        resid_pdrop=0.1, # llama doesn't have this
-        partial_rotary_factor=0.5,
+        resid_pdrop=0.1,           # llama doesn't have this
+        partial_rotary_factor=0.5, # important difference between llama and phi
         **kwargs,
     ):
         self.vocab_size = vocab_size
-        self.max_position_embeddings = max_position_embeddings
         self.hidden_size = hidden_size
         self.num_hidden_layers = num_hidden_layers
         self.num_attention_heads = num_attention_heads
         self.num_key_value_heads = num_key_value_heads
         self.hidden_act = hidden_act
-        self.initializer_range = initializer_range
         self.layer_norm_eps = layer_norm_eps
-        self.pretraining_tp = pretraining_tp
-        self.use_cache = use_cache
-        self.rope_scaling = rope_scaling
         self.rope_theta = rope_theta
         self.resid_pdrop = resid_pdrop
         self.partial_rotary_factor = partial_rotary_factor
@@ -116,16 +106,16 @@ class FlashPhiAttention(torch.nn.Module):
         self.hidden_size = config.hidden_size
         self.head_size = self.hidden_size // self.num_heads
 
+        self.softmax_scale = self.head_size**-0.5
+        self.rotary_dim = int(config.partial_rotary_factor * self.head_size)
+
         self.rotary_emb = PositionRotaryEmbedding.static(
             config=config,
-            dim=self.num_heads,
+            dim=self.rotary_dim,
             base=config.rope_theta,
             device=weights.device,
         )
 
-        self.softmax_scale = self.head_size**-0.5
-        self.rotary_dim = int(config.partial_rotary_factor * self.head_size)
-
         if self.num_heads % weights.process_group.size() != 0:
             raise ValueError(
                 f"`num_heads` must be divisible by `num_shards` (got `num_heads`: {self.num_heads} "