Move to in-library checkpoint

configuration_jamba.py

@@ -26,9 +26,9 @@ class JambaConfig(PretrainedConfig):
     r"""
     This is the configuration class to store the configuration of a [`JambaModel`]. It is used to instantiate a
     Jamba model according to the specified arguments, defining the model architecture. Instantiating a configuration
-    with the defaults will yield a similar configuration to that of the jamba-small architecture.
+    with the defaults will yield a similar configuration to that of the Jamba-v0.1 model.
 
-    [ai21labs/jamba-small](https://huggingface.co/ai21labs/Jamba-v0.1)
+    [ai21labs/Jamba-v0.1](https://huggingface.co/ai21labs/Jamba-v0.1)
 
     Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
     documentation from [`PretrainedConfig`] for more information.
@@ -65,12 +65,12 @@ class JambaConfig(PretrainedConfig):
         use_cache (`bool`, *optional*, defaults to `True`):
             Whether or not the model should return the last key/values attentions (not used by all models). Only
             relevant if `config.is_decoder=True`.
-        calc_logits_for_entire_prompt (`bool`, *optional*, defaults to `False`):
-            Whether or not to calculate logits for entire prompt during generation. If `False`, only the logits of the
-            last prompt token will be calculated, which are the only logits needed for generation. For long sequences,
-            the logits for the entire sequence may use a lot of memory so setting `calc_logits_for_entire_prompt=False`
-            will reduce memory footprint significantly.
-            Note: some generation features may not be available if this is set to `False`.
+        num_logits_to_keep (`int` or `None`, *optional*, defaults to 1):
+            Number of prompt logits to calculate during generation. If `None`, all logits will be calculated. If an
+            integer value, only last `num_logits_to_keep` logits will be calculated. Default is 1 because only the
+            logits of the last prompt token are needed for generation. For long sequences, the logits for the entire
+            sequence may use a lot of memory so, setting `num_logits_to_keep=1` will reduce memory footprint
+            significantly.
         output_router_logits (`bool`, *optional*, defaults to `False`):
             Whether or not the router logits should be returned by the model. Enabling this will also
             allow the model to output the auxiliary loss. See [here]() for more details
@@ -84,7 +84,7 @@ class JambaConfig(PretrainedConfig):
             The id of the "end-of-sequence" token.
         sliding_window (`int`, *optional*):
             Sliding window attention window size. If not specified, will default to `None`.
-        n_ctx (`int`, *optional*, defaults to 262144):
+        max_position_embeddings (`int`, *optional*, defaults to 262144):
             This value doesn't have any real effect. The maximum sequence length that this model is intended to be
             used with. It can be used with longer sequences, but performance may degrade.
         attention_dropout (`float`, *optional*, defaults to 0.0):
@@ -118,8 +118,6 @@ class JambaConfig(PretrainedConfig):
             Flag indicating whether or not to use bias in the convolution layer of the mamba mixer block.
         mamba_proj_bias (`bool`, *optional*, defaults to `False`):
             Flag indicating whether or not to use bias in the input and output projections (["in_proj", "out_proj"]) of the mamba mixer block
-        mamba_inner_layernorms (`bool`, *optional*, defaults to `True`):
-            Flag indicating whether or not to apply layernorms to internal mamba activations
 
     """
 
@@ -139,14 +137,14 @@ class JambaConfig(PretrainedConfig):
             initializer_range=0.02,
             rms_norm_eps=1e-6,
             use_cache=True,
-            calc_logits_for_entire_prompt=False,
+            num_logits_to_keep=1,
             output_router_logits=False,
             router_aux_loss_coef=0.001,
             pad_token_id=0,
             bos_token_id=1,
             eos_token_id=2,
             sliding_window=None,
-            n_ctx=262144,
+            max_position_embeddings=262144,
             attention_dropout=0.0,
             num_experts_per_tok=2,
             num_experts=16,
@@ -161,7 +159,6 @@ class JambaConfig(PretrainedConfig):
             mamba_dt_rank="auto",
             mamba_conv_bias=True,
             mamba_proj_bias=False,
-            mamba_inner_layernorms=True,
             **kwargs,
     ):
         self.vocab_size = vocab_size
@@ -171,7 +168,7 @@ class JambaConfig(PretrainedConfig):
         self.num_hidden_layers = num_hidden_layers
         self.num_attention_heads = num_attention_heads
         self.sliding_window = sliding_window
-        self.n_ctx = n_ctx
+        self.max_position_embeddings = max_position_embeddings
         self.attention_dropout = attention_dropout
 
         # for backward compatibility
@@ -184,7 +181,7 @@ class JambaConfig(PretrainedConfig):
         self.rms_norm_eps = rms_norm_eps
 
         self.use_cache = use_cache
-        self.calc_logits_for_entire_prompt = calc_logits_for_entire_prompt
+        self.num_logits_to_keep = num_logits_to_keep
         self.output_router_logits = output_router_logits
         self.router_aux_loss_coef = router_aux_loss_coef
 
@@ -202,7 +199,6 @@ class JambaConfig(PretrainedConfig):
         self.mamba_dt_rank = math.ceil(self.hidden_size / 16) if mamba_dt_rank == "auto" else mamba_dt_rank
         self.mamba_conv_bias = mamba_conv_bias
         self.mamba_proj_bias = mamba_proj_bias
-        self.mamba_inner_layernorms = mamba_inner_layernorms
 
         super().__init__(
             pad_token_id=pad_token_id,
@@ -211,3 +207,17 @@ class JambaConfig(PretrainedConfig):
             tie_word_embeddings=tie_word_embeddings,
             **kwargs,
         )
+
+    @property
+    def layers_block_type(self):
+        return [
+            "attention" if i % self.attn_layer_period == self.attn_layer_offset else "mamba"
+            for i in range(self.num_hidden_layers)
+        ]
+
+    @property
+    def layers_num_experts(self):
+        return [
+            self.num_experts if i % self.expert_layer_period == self.expert_layer_offset else 1
+            for i in range(self.num_hidden_layers)
+        ]