adding multi token input and correct cache handling in ssm step

llms/mlx_lm/models/mamba2.py

@@ -27,10 +27,10 @@ class ModelArgs(BaseModelArgs):
     time_step_max: float
     time_step_floor: float
     rescale_prenorm_residual: bool
-    use_cache: bool
     rms_norm: bool
     chunk_size: int
     tie_word_embeddings: bool
+    use_cache: bool = True
     time_step_limit: Tuple[float, float] = field(default_factory=lambda: (0.0, float("inf")))
     time_step_rank: Union[int, str] = "auto"
     model_type: str = "mamba2"
@@ -58,6 +58,29 @@ class MambaRMSNormGated(nn.Module):
         return self.weight * hidden_states
     
 
+def silu(x):
+    return x * mx.sigmoid(x)
+
+def ssd(x, A, B, C, chunk_size):
+    batch, seqlen, nheads, dim = x.shape
+    B = mx.expand_dims(B, axis=2)
+    C = mx.expand_dims(C, axis=2)
+    
+    state = mx.zeros((batch, nheads, dim, B.shape[-1]))
+    outputs = []
+    
+    for i in range(0, seqlen, chunk_size):
+        chunk = slice(i, min(i + chunk_size, seqlen))
+        dA = mx.exp(mx.expand_dims(A[chunk], axis=0))
+        
+        dBx = mx.einsum('blhp,bln->bhpn', x[:, chunk], B[:, chunk])
+        state = state * mx.expand_dims(dA, axis=-1) + dBx
+        y = mx.einsum('bhpn,bln->blhp', state, C[:, chunk])
+        outputs.append(y)
+    
+    return mx.concatenate(outputs, axis=1), state
+    
+
 class DepthWiseConv1d(nn.Module):
     def __init__(self, in_channels, out_channels, kernel_size, bias=True, groups=None, padding=0):
         super().__init__()
@@ -67,127 +90,142 @@ class DepthWiseConv1d(nn.Module):
         self.padding = padding
         self.groups = groups if groups is not None else in_channels
         
-        # Ensure in_channels and out_channels are the same for depthwise conv
+        assert in_channels == out_channels, "In and out channels must be same for depthwise convolution"
-        assert in_channels == out_channels, "In and out channels must be the same for depthwise convolution"
-        # Ensure groups is equal to in_channels for depthwise conv
         assert self.groups == in_channels, "Groups must be equal to in_channels for depthwise convolution"
         
-        # Initialize weight with shape (out_channels, kernel_size, 1)
+        # Initialize with shape (channels, 1, kernel_size) to match pretrained weights
-        self.weight = mx.random.normal((out_channels, kernel_size, 1))
+        self.weight = mx.random.normal((in_channels, 1, kernel_size))
         self.bias = mx.zeros((out_channels,)) if bias else None
 
-    def __call__(self, x, cache=None):
+    def __call__(self, x: mx.array, cache=None, cache_idx: int = 0) -> mx.array:
         B, L, C = x.shape
-        _, K, _ = self.weight.shape
+        K = self.kernel_size
 
+        # Handle padding and caching
         if cache is not None:
-            x = mx.concatenate([cache, x], axis=1)
+            conv_cache = cache[cache_idx]
+            if conv_cache is not None:
+                x = mx.concatenate([conv_cache, x], axis=1)
+                L = x.shape[1]  # Update L after concatenation
         else:
-            x = mx.pad(x, [(0, 0), (K - 1, 0), (0, 0)])
+            pad_left = K - 1
+            x = mx.pad(x, [(0, 0), (pad_left, 0), (0, 0)])
+            L = x.shape[1]  # Update L after padding
 
-        y = mx.conv_general(x, self.weight, groups=self.groups)
+        # Implement depthwise convolution manually for each channel
+        outputs = []
+        for c in range(C):
+            # Extract single channel and reshape for 1D convolution
+            x_c = x[:, :, c]  # Shape: [B, L]
+            x_c = mx.expand_dims(x_c, axis=1)  # Shape: [B, 1, L]
+            
+            # Extract and ensure filter is 3D
+            w_c = self.weight[c]  # Shape: [1, kernel_size] or [1, 1, kernel_size]
+            if w_c.ndim == 2:
+                w_c = mx.expand_dims(w_c, axis=0)  # Shape: [1, 1, kernel_size]
+            elif w_c.ndim == 1:
+                w_c = mx.expand_dims(mx.expand_dims(w_c, axis=0), axis=0)
+                
+            # For inference mode (single token), adjust the input
+            if L < K:
+                # Pad input to match kernel size
+                pad_size = K - L
+                x_c = mx.pad(x_c, [(0, 0), (0, 0), (pad_size, 0)])
+            
+            # Apply 1D convolution for this channel
+            y_c = mx.conv_general(
+                x_c,
+                w_c,
+                stride=1,
+                padding=0  # We've already handled padding
+            )
             
             if self.bias is not None:
-            y = y + self.bias
+                y_c = y_c + self.bias[c]
             
-        return y, x[:, -K + 1 :, :]
+            outputs.append(mx.squeeze(y_c, axis=1))  # Shape: [B, 1]
+        
+        # Stack all channel outputs
+        y = mx.stack(outputs, axis=-1)  # Shape: [B, L', C]
+        
+        if cache is not None:
+            # Update cache with the most recent K-1 tokens
+            cache[cache_idx] = x[:, -(K-1):, :] if L >= K else x
+
+        return y
     
 
 class Mamba2Block(nn.Module):
     def __init__(self, args: ModelArgs):
         super().__init__()
         self.args = args
-        self.intermediate_size = args.intermediate_size
-        self.time_step_rank = args.time_step_rank
-        self.conv_kernel_size = args.conv_kernel
-        self.hidden_size = args.hidden_size
-        self.state_size = args.state_size
-        self.num_heads = args.num_heads
-        self.head_dim = args.hidden_size // args.num_heads
-        self.n_groups = args.n_groups
         
-        # projection_size = 2 * args.intermediate_size + 2 * args.n_groups * args.state_size + args.num_heads
+        d_in_proj = 2 * args.intermediate_size + 2 * args.state_size + args.num_heads
-        projection_size = 2 * args.intermediate_size + 2 * args.state_size + args.num_heads
+        self.in_proj = nn.Linear(args.hidden_size, d_in_proj, bias=args.use_bias)
-        self.in_proj = nn.Linear(
-            args.hidden_size,
-            projection_size,
-            bias=args.use_bias
-        )
 
-        # self.conv_dim = args.intermediate_size + 2 * args.n_groups * args.state_size
+        conv_dim = args.intermediate_size + 2 * args.state_size
-        self.conv_dim = args.intermediate_size + 2 * args.state_size
         self.conv1d = DepthWiseConv1d(
-            in_channels=self.conv_dim,
+            in_channels=conv_dim,
-            out_channels=self.conv_dim,
+            out_channels=conv_dim,
             kernel_size=args.conv_kernel,
+            groups=conv_dim,
             bias=args.use_conv_bias,
-            groups=self.conv_dim,
             padding=args.conv_kernel - 1
         )
 
-        self.A_log = mx.zeros(args.num_heads)
+        self.dt_bias = mx.random.normal((args.num_heads,)) * args.initializer_range
-        self.D = mx.ones((args.num_heads,))
+        self.A_log = mx.random.normal((args.num_heads,)) * args.initializer_range
-        self.dt_bias = mx.zeros(args.num_heads)
+        self.D = mx.random.normal((args.num_heads,)) * args.initializer_range
 
-        self.out_proj = nn.Linear(args.intermediate_size, args.hidden_size, bias=args.use_bias)
         self.norm = MambaRMSNormGated(args.intermediate_size, eps=args.layer_norm_epsilon)
+        self.out_proj = nn.Linear(args.intermediate_size, args.hidden_size, bias=args.use_bias)
 
-    def _ssd(self, x, A, B, C, chunk_size):
+        if args.rescale_prenorm_residual:
-        batch, seq_len, nheads, head_dim = x.shape
+            layer_scale = math.sqrt(1.0 / args.num_hidden_layers)
-        n_state = B.shape[-1]
+            self.out_proj.weight = self.out_proj.weight * layer_scale
 
-        h = mx.zeros((batch, nheads, head_dim, n_state))
+    def __call__(self, u: mx.array, cache = None):
-        ys = []
+        if cache is not None and self.args.use_cache:
-        
+            return self.step(u, cache)
-        for i in range(0, seq_len, chunk_size):
-            chunk_size_i = min(chunk_size, seq_len - i)
-            xi = x[:, i:i + chunk_size_i]
-            Bi = B[:, i:i + chunk_size_i]
-            Ci = C[:, i:i + chunk_size_i]
-            
-            for t in range(chunk_size_i):
-                h = h * mx.exp(A)[:, None, None]
-                h = h + mx.expand_dims(Bi[:, t], -2) * mx.expand_dims(xi[:, t], -1)
-                y = mx.sum(h * mx.expand_dims(Ci[:, t], -2), axis=-1)
-                ys.append(y)
-        
-        y = mx.stack(ys, axis=1)
-        return y, h
-
-    def __call__(self, x: mx.array, cache) -> mx.array:
-        if cache is not None:
-            return self.step(x, cache)
 
         A = -mx.exp(self.A_log)
         zxbcdt = self.in_proj(u)
         
-        z, xBC, dt = mx.split(
+        splits = [
-            zxbcdt,
+            self.args.intermediate_size,
-            [
+            self.args.intermediate_size + 2 * self.args.state_size,
-                self.args.d_inner,
+            self.args.num_heads,
-                self.args.d_inner + 2 * self.args.d_state,
+        ]
-                self.args.nheads,
+
-            ],
+        z, xBC, dt = mx.split(zxbcdt, splits, axis=-1)
-            axis=-1,
+        
+        dt = mx.clip(
+            nn.softplus(dt + self.dt_bias),
+            self.args.time_step_min,
+            self.args.time_step_max
         )
         
-        dt = mx.softplus(dt + self.dt_bias)
+        dt = mx.maximum(dt, self.args.time_step_floor)
 
-        # Use the custom DepthWiseConv1d with cache
+        xBC = silu(self.conv1d(xBC))
-        xBC = self.conv1d(xBC, cache, cache_idx=0)
-        xBC = mx.sigmoid(xBC) * xBC  # SiLU activation
 
-        x, B, C = mx.split(
+        xBC_parts = mx.split(
             xBC,
-            [self.args.d_inner, self.args.d_state, self.args.d_state],
+            [self.args.intermediate_size, self.args.state_size, self.args.state_size],
             axis=-1
         )
 
-        x = self._reshape_heads(x, True)
+        x = xBC_parts[0]
-        B = mx.expand_dims(B, axis=2)
+        B = xBC_parts[1]
-        C = mx.expand_dims(C, axis=2)
+        C = xBC_parts[2]
         
-        y, ssm_state = self._ssd(
+        # Replace rearrange with reshape and transpose
+        b, l, hp = x.shape
+        h = self.args.num_heads
+        p = hp // h
+        x = mx.reshape(x, (b, l, h, p))
+        
+        y, ssm_state = ssd(
             x * mx.expand_dims(dt, -1),
             A * dt,
             B,
@@ -196,61 +234,127 @@ class Mamba2Block(nn.Module):
         )
         
         y = y + x * mx.expand_dims(self.D, -1)
-        y = self._reshape_heads(y, False)
+        # Replace rearrange with reshape
-        y = self.norm(y, z)
+        y = mx.reshape(y, (b, l, h * p))
+        
+        y = self.norm(y + z)
         y = self.out_proj(y)
 
-        if cache is not None:
+        if cache is not None and self.args.use_cache:
             cache[1] = ssm_state
 
+        if self.args.residual_in_fp32:
+            y = mx.cast(y, mx.float32)
+
         return y
 
-    def step(self, x: mx.array, cache) -> mx.array:
+    def step(self, u: mx.array, cache: MambaCache):
-        """Single inference step"""
+        batch_size = u.shape[0]
-        assert x.shape[1] == 1, "Only one token can be decoded per inference step"
+        seq_len = u.shape[1]
+        outputs = []
 
-        zxbcdt = self.in_proj(mx.squeeze(x, 1))
+        # Initialize SSM state if needed
-        z, xBC, dt = mx.split(
+        if cache[1] is None:
-            zxbcdt,
+            cache[1] = mx.zeros((
-            [
+                batch_size,
-                self.args.d_inner,
+                self.args.num_heads,
-                self.args.d_inner + 2 * self.args.d_state,
+                self.args.head_dim,
-                self.args.nheads,
+                self.args.state_size
-            ],
+            ))
-            axis=-1,
+
+        for pos in range(seq_len):
+            # Get single token
+            u_t = u[:, pos:pos+1, :]
+
+            # Project input
+            zxbcdt = self.in_proj(u_t)
+            
+            # Calculate sizes
+            d_model = self.args.intermediate_size
+            d_state = self.args.state_size
+            n_heads = self.args.num_heads
+            d_head = self.args.head_dim
+            
+            # Correct splits for z, xBC, dt
+            splits = [
+                d_model,                    # z size
+                d_model + 2 * d_state,      # xBC size (delta, B, C)
+                n_heads                     # dt size
+            ]
+            
+            # Split the projected input
+            z = zxbcdt[:, :, :splits[0]]
+            xBC = zxbcdt[:, :, splits[0]:splits[0] + splits[1]]
+            dt = zxbcdt[:, :, -splits[2]:]  # Take last n_heads elements
+
+            # Process dt
+            dt = mx.reshape(dt, (batch_size, n_heads))
+            dt = mx.clip(
+                nn.softplus(dt + self.dt_bias),
+                self.args.time_step_min,
+                self.args.time_step_max
             )
+            dt = mx.maximum(dt, self.args.time_step_floor)
 
-        # Use the custom DepthWiseConv1d with cache
+            # Process convolution
-        xBC = self.conv1d(xBC, cache, cache_idx=0)
+            xBC = self.conv1d(xBC, cache=cache, cache_idx=0)
-        xBC = mx.sigmoid(xBC) * xBC  # SiLU activation
+            xBC = silu(xBC)
 
-        x, B, C = mx.split(
+            # Split convolved xBC into x, B, C
-            xBC,
+            x = xBC[:, :, :d_model]
-            [self.args.d_inner, self.args.d_state, self.args.d_state],
+            B = xBC[:, :, d_model:d_model + d_state]
-            axis=-1
+            C = xBC[:, :, -d_state:]
-        )
+            
+            # Reshape x into (batch, heads, dim)
+            x = mx.reshape(x, (batch_size, 1, n_heads, d_head))
+            x = mx.squeeze(x, axis=1)  # (batch, heads, dim)
+            
+            # Reshape B into (batch, heads, dim, state)
+            B = mx.reshape(B, (batch_size, 1, d_state))
+            B = mx.broadcast_to(B, (batch_size, n_heads, d_state))
+            B = mx.expand_dims(B, axis=2)  # (batch, heads, 1, state)
+
+            # Reshape C for later use
+            C = mx.reshape(C, (batch_size, 1, d_state))
+            C = mx.broadcast_to(C, (batch_size, n_heads, d_state))
+            C = mx.expand_dims(C, axis=3)  # (batch, heads, state, 1)
+
+            # Compute SSM updates
             A = -mx.exp(self.A_log)
+            dA = mx.exp(dt * mx.expand_dims(A, 0))
+            dA = mx.expand_dims(mx.expand_dims(dA, -1), -1)  # (batch, heads, 1, 1)
 
-        dt = mx.softplus(dt + self.dt_bias)
+            # Prepare x for Bx computation
-        dA = mx.exp(dt * A)
+            x = mx.expand_dims(x, axis=3)  # (batch, heads, dim, 1)
             
-        x = mx.reshape(x, (-1, self.args.nheads, self.args.headdim))
+            # Compute dBx with proper broadcasting
+            dBx = mx.matmul(x, B)  # (batch, heads, dim, state)
 
-        ssm_state = cache[1]
+            # Update state
-        dBx = mx.expand_dims(dt, -1) * mx.expand_dims(B, 1) * mx.expand_dims(x, -1)
+            ssm_state = cache[1]  # (batch, heads, dim, state)
-        ssm_state = ssm_state * mx.expand_dims(mx.expand_dims(dA, -1), -1) + dBx
+            ssm_state = ssm_state * dA + dBx
-        
-        y = mx.sum(ssm_state * mx.expand_dims(mx.expand_dims(C, 1), 1), axis=-1)
-        y = y + mx.expand_dims(self.D, -1) * x
-        y = mx.reshape(y, (-1, self.args.nheads * self.args.headdim))
-        
-        y = self.norm(y, z)
-        y = self.out_proj(y)
-
-        # Update SSM state in cache
             cache[1] = ssm_state
 
-        return mx.expand_dims(y, 1)
+            # Compute output
+            y = mx.matmul(ssm_state, C)  # (batch, heads, dim, 1)
+            y = mx.squeeze(y, axis=-1)    # (batch, heads, dim)
+
+            # Add skip connection with D
+            y = y + x[:, :, :, 0] * mx.expand_dims(self.D, -1)
+
+            # Reshape to original dimensions
+            y = mx.reshape(y, (batch_size, 1, n_heads * d_head))
+
+            # Apply norm and output projection
+            y = self.norm(y + z)
+            y = self.out_proj(y)
+
+            if self.args.residual_in_fp32:
+                y.astype(mx.float32)
+
+            outputs.append(y)
+
+        return mx.concatenate(outputs, axis=1)
 
 
 class ResidualBlock(nn.Module):
@@ -287,7 +391,6 @@ class Model(nn.Module):
         self.model_type = args.model_type
         
         self.backbone = Mamba2(args)
-        # self.norm_f = nn.RMSNorm(args.hidden_size, eps=args.layer_norm_epsilon)
 
         if not args.tie_word_embeddings:
             self.lm_head = nn.Linear(args.hidden_size, args.vocab_size, bias=False)
@@ -302,17 +405,26 @@ class Model(nn.Module):
         else:
             logits = self.lm_head(x)
 
+        print('ouput')
         return logits
 
-    def sanitize(self, weights):
-        for k, v in weights.items():
-            if "conv1d.weight" in k and v.ndim == 3:
-                weights[k] = v.moveaxis(2, 1)
-        return weights
-
     def make_cache(self):
         return [MambaCache() for _ in range(len(self.layers))]
     
+    def sanitize(self, weights):
+        sanitized = {}
+        for k, v in weights.items():
+            if "conv1d.weight" in k:
+                # Ensure weights are in correct shape (channels, 1, kernel_size)
+                if v.ndim == 2:
+                    v = mx.expand_dims(v, axis=1)
+                elif v.ndim == 1:
+                    v = mx.expand_dims(mx.expand_dims(v, axis=0), axis=0)
+                sanitized[k] = v
+            else:
+                sanitized[k] = v
+        return sanitized
+
     @property
     def layers(self):
         return self.backbone.layers

llms/mlx_lm/tuner/utils.py

@@ -146,6 +146,8 @@ def linear_to_lora_layers(
     elif model.model_type == "mamba2":
         keys = set(
             [
+                "mixer.in_proj",
+                "mixer.out_proj",
             ]
         )
     else: