Refactor KVCache and add stream generate (#53)

* refactor KVcache and add stream generate

* update KV Cache

* fix tests and update generate

* temporary fix to deepseek and paligema

* refactor chat_template and fix idefics KV cache

* remove tuple and update model

mlx_vlm/generate.py

@@ -3,10 +3,16 @@
 
 import mlx.core as mx
 
-from .prompt_utils import get_message_json
+from .prompt_utils import apply_chat_template
 from .utils import generate, get_model_path, load, load_config, load_image_processor
 
-MODEL_TYPE = ""
+DEFAULT_MODEL_PATH = "mlx-community/nanoLLaVA-1.5-8bit"
+DEFAULT_IMAGE = "http://images.cocodataset.org/val2017/000000039769.jpg"
+DEFAULT_PROMPT = "What are these?"
+DEFAULT_MAX_TOKENS = 100
+DEFAULT_TEMP = 0.5
+DEFAULT_TOP_P = 1.0
+DEFAULT_SEED = 0
 
 
 def parse_arguments():
@@ -16,36 +22,31 @@ def parse_arguments():
     parser.add_argument(
         "--model",
         type=str,
-        default="qnguyen3/nanoLLaVA",
+        default=DEFAULT_MODEL_PATH,
         help="The path to the local model directory or Hugging Face repo.",
     )
     parser.add_argument(
         "--image",
         type=str,
-        default="http://images.cocodataset.org/val2017/000000039769.jpg",
+        default=DEFAULT_IMAGE,
         help="URL or path of the image to process.",
     )
     parser.add_argument(
         "--prompt",
         type=str,
-        default="What are these?",
+        default=DEFAULT_PROMPT,
         help="Message to be processed by the model.",
     )
     parser.add_argument(
         "--max-tokens",
         type=int,
-        default=100,
+        default=DEFAULT_MAX_TOKENS,
         help="Maximum number of tokens to generate.",
     )
     parser.add_argument(
-        "--temp", type=float, default=0.3, help="Temperature for sampling."
-    )
-    parser.add_argument(
-        "--verbose",
-        type=bool,
-        help="Detailed output.",
-        default=True,
+        "--temp", type=float, default=DEFAULT_TEMP, help="Temperature for sampling."
     )
+    parser.add_argument("--verbose", action="store_false", help="Detailed output.")
     return parser.parse_args()
 
 
@@ -57,38 +58,14 @@ def get_model_and_processors(model_path):
     return model, processor, image_processor, config
 
 
-def sample(logits, temperature=0.0):
-    if temperature == 0:
-        return mx.argmax(logits, axis=-1)
-    else:
-        return mx.random.categorical(logits * (1 / temperature))
-
-
 def main():
     args = parse_arguments()
     model, processor, image_processor, config = get_model_and_processors(args.model)
 
     prompt = codecs.decode(args.prompt, "unicode_escape")
 
-    if "chat_template" in processor.__dict__.keys():
-        prompt = processor.apply_chat_template(
-            [get_message_json(config["model_type"], prompt)],
-            tokenize=False,
-            add_generation_prompt=True,
-        )
-
-    elif "tokenizer" in processor.__dict__.keys():
-        if model.config.model_type != "paligemma":
-            prompt = processor.tokenizer.apply_chat_template(
-                [get_message_json(config["model_type"], prompt)],
-                tokenize=False,
-                add_generation_prompt=True,
-            )
-
-    else:
-        ValueError(
-            "Error: processor does not have 'chat_template' or 'tokenizer' attribute."
-        )
+    if model.config.model_type != "paligemma":
+        prompt = apply_chat_template(processor, config, prompt)
 
     output = generate(
         model,

mlx_vlm/models/base.py

@@ -1,6 +1,7 @@
 from abc import ABC, abstractmethod
 from typing import Dict
 
+import mlx.core as mx
 from PIL import Image
 from transformers.image_processing_utils import get_size_dict
 from transformers.image_utils import ChannelDimension, PILImageResampling
@@ -49,3 +50,41 @@ def __init__(
     @abstractmethod
     def preprocess(self, images):
         pass
+
+
+class KVCache:
+
+    def __init__(self, head_dim, n_kv_heads):
+        self.n_kv_heads = n_kv_heads
+        if isinstance(head_dim, int):
+            self.k_head_dim = self.v_head_dim = head_dim
+        elif isinstance(head_dim, tuple) and len(head_dim) == 2:
+            self.k_head_dim, self.v_head_dim = head_dim
+        else:
+            raise ValueError("head_dim must be an int or a tuple of two ints")
+        self.keys = None
+        self.values = None
+        self.offset = 0
+        self.step = 256
+
+    def update_and_fetch(self, keys, values):
+        prev = self.offset
+        if self.keys is None or (prev + keys.shape[2]) > self.keys.shape[2]:
+            n_steps = (self.step + keys.shape[2] - 1) // self.step
+            k_shape = (1, self.n_kv_heads, n_steps * self.step, self.k_head_dim)
+            v_shape = (1, self.n_kv_heads, n_steps * self.step, self.v_head_dim)
+            new_k = mx.zeros(k_shape, keys.dtype)
+            new_v = mx.zeros(v_shape, values.dtype)
+            if self.keys is not None:
+                if prev % self.step != 0:
+                    self.keys = self.keys[..., :prev, :]
+                    self.values = self.values[..., :prev, :]
+                self.keys = mx.concatenate([self.keys, new_k], axis=2)
+                self.values = mx.concatenate([self.values, new_v], axis=2)
+            else:
+                self.keys, self.values = new_k, new_v
+
+        self.offset += keys.shape[2]
+        self.keys[..., prev : self.offset, :] = keys
+        self.values[..., prev : self.offset, :] = values
+        return self.keys[..., : self.offset, :], self.values[..., : self.offset, :]

mlx_vlm/models/idefics2/idefics2.py

@@ -254,10 +254,10 @@ def __call__(
         self, input_ids: mx.array, pixel_values: mx.array, mask: mx.array, cache=None
     ):
         input_embeddings = self.get_input_embeddings(input_ids, pixel_values)
-        logits, cache = self.language_model(
+        logits = self.language_model(
             inputs=input_ids, cache=cache, inputs_embeds=input_embeddings
         )
-        return logits, cache
+        return logits
 
     @staticmethod
     def from_pretrained(path_or_hf_repo: str):

mlx_vlm/models/idefics2/language.py

@@ -37,14 +37,14 @@ def __post_init__(self):
 
 
 class Attention(nn.Module):
-    def __init__(self, args: TextConfig):
+    def __init__(self, config: TextConfig):
         super().__init__()
 
-        dim = args.hidden_size
-        self.n_heads = n_heads = args.num_attention_heads
-        self.n_kv_heads = n_kv_heads = args.num_key_value_heads
+        dim = config.hidden_size
+        self.n_heads = n_heads = config.num_attention_heads
+        self.n_kv_heads = n_kv_heads = config.num_key_value_heads
 
-        head_dim = args.hidden_size // n_heads
+        head_dim = config.hidden_size // n_heads
         self.scale = head_dim**-0.5
 
         self.q_proj = nn.Linear(dim, n_heads * head_dim, bias=False)
@@ -54,8 +54,8 @@ def __init__(self, args: TextConfig):
 
         self.rope = nn.RoPE(
             head_dim,
-            traditional=args.rope_traditional,
-            base=args.rope_theta,
+            traditional=config.rope_traditional,
+            base=config.rope_theta,
         )
 
     def __call__(
@@ -74,11 +74,9 @@ def __call__(
         values = values.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)
 
         if cache is not None:
-            key_cache, value_cache = cache
-            queries = self.rope(queries, offset=key_cache.shape[2])
-            keys = self.rope(keys, offset=key_cache.shape[2])
-            keys = mx.concatenate([key_cache, keys], axis=2)
-            values = mx.concatenate([value_cache, values], axis=2)
+            queries = self.rope(queries, offset=cache.offset)
+            keys = self.rope(keys, offset=cache.offset)
+            keys, values = cache.update_and_fetch(keys, values)
         else:
             queries = self.rope(queries)
             keys = self.rope(keys)
@@ -87,7 +85,7 @@ def __call__(
             queries, keys, values, scale=self.scale, mask=mask
         )
         output = output.transpose(0, 2, 1, 3).reshape(B, L, -1)
-        return self.o_proj(output), (keys, values)
+        return self.o_proj(output)
 
 
 class MLP(nn.Module):
@@ -102,45 +100,45 @@ def __call__(self, x) -> mx.array:
 
 
 class TransformerBlock(nn.Module):
-    def __init__(self, args: TextConfig):
+    def __init__(self, config: TextConfig):
         super().__init__()
-        self.num_attention_heads = args.num_attention_heads
-        self.hidden_size = args.hidden_size
-        self.self_attn = Attention(args)
-        self.mlp = MLP(args.hidden_size, args.intermediate_size)
-        self.input_layernorm = nn.RMSNorm(args.hidden_size, eps=args.rms_norm_eps)
+        self.num_attention_heads = config.num_attention_heads
+        self.hidden_size = config.hidden_size
+        self.self_attn = Attention(config)
+        self.mlp = MLP(config.hidden_size, config.intermediate_size)
+        self.input_layernorm = nn.RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.post_attention_layernorm = nn.RMSNorm(
-            args.hidden_size, eps=args.rms_norm_eps
+            config.hidden_size, eps=config.rms_norm_eps
         )
-        self.args = args
+        self.config = config
 
     def __call__(
         self,
         x: mx.array,
         mask: Optional[mx.array] = None,
         cache: Optional[Tuple[mx.array, mx.array]] = None,
     ) -> mx.array:
-        r, cache = self.self_attn(self.input_layernorm(x), mask, cache)
+        r = self.self_attn(self.input_layernorm(x), mask, cache)
         h = x + r
         r = self.mlp(self.post_attention_layernorm(h))
         out = h + r
-        return out, cache
+        return out
 
 
 class LanguageModel(nn.Module):
-    def __init__(self, args: TextConfig):
+    def __init__(self, config: TextConfig):
         super().__init__()
-        self.args = args
-        self.model_type = args.model_type
-        self.vocab_size = args.vocab_size
-        self.num_hidden_layers = args.num_hidden_layers
+        self.config = config
+        self.model_type = config.model_type
+        self.vocab_size = config.vocab_size
+        self.num_hidden_layers = config.num_hidden_layers
         assert self.vocab_size > 0
-        self.embed_tokens = nn.Embedding(args.vocab_size, args.hidden_size)
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
         self.layers = [
-            TransformerBlock(args=args) for _ in range(args.num_hidden_layers)
+            TransformerBlock(config=config) for _ in range(config.num_hidden_layers)
         ]
-        self.norm = nn.RMSNorm(args.hidden_size, eps=args.rms_norm_eps)
-        self.lm_head = nn.Linear(args.hidden_size, args.vocab_size, bias=False)
+        self.norm = nn.RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
 
     def __call__(
         self,
@@ -163,10 +161,10 @@ def __call__(
         if cache is None:
             cache = [None] * len(self.layers)
 
-        for e, layer in enumerate(self.layers):
-            h, cache[e] = layer(h, mask, cache[e])
+        for layer, c in zip(self.layers, cache):
+            h = layer(h, mask, c)
 
-        return self.lm_head(self.norm(h)), cache
+        return self.lm_head(self.norm(h))
 
     def sanitize(self, weights):
         # Remove unused precomputed rotary freqs
@@ -177,3 +175,11 @@ def sanitize(self, weights):
     @property
     def layers(self):
         return self.model.layers
+
+    @property
+    def head_dim(self):
+        return self.config.hidden_size // self.config.num_attention_heads
+
+    @property
+    def n_kv_heads(self):
+        return self.config.num_key_value_heads