From b10508fd6fa108d90cfb1ec9a53e33a15eadc459 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Thu, 15 May 2025 17:15:34 +0000
Subject: [PATCH 01/13] Onboarding qwen3moe

Signed-off-by: Dipankar Sarkar <quic_dipankar@quicinc.com>
---
 .../transformers/models/pytorch_transforms.py |  25 +
 .../models/qwen3_moe/modeling_qwen3_moe.py    | 599 ++++++++++++++++++
 2 files changed, 624 insertions(+)
 create mode 100644 QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py

diff --git a/QEfficient/transformers/models/pytorch_transforms.py b/QEfficient/transformers/models/pytorch_transforms.py
index 4297723c3..9f0f2cd3c 100644
--- a/QEfficient/transformers/models/pytorch_transforms.py
+++ b/QEfficient/transformers/models/pytorch_transforms.py
@@ -115,6 +115,15 @@
     Qwen2Model,
     Qwen2RMSNorm,
 )
+from transformers.models.qwen3_moe.modeling_qwen3_moe import (
+    Qwen3MoeAttention,
+    Qwen3MoeDecoderLayer,
+    Qwen3MoeForCausalLM,
+    Qwen3MoeModel,
+    Qwen3MoeRMSNorm,
+    Qwen3MoeRotaryEmbedding,
+    Qwen3MoeSparseMoeBlock,
+)
 from transformers.models.starcoder2.modeling_starcoder2 import (
     Starcoder2Attention,
     Starcoder2DecoderLayer,
@@ -251,6 +260,14 @@
     QEffQwen2ForCausalLM,
     QEffQwen2Model,
 )
+from QEfficient.transformers.models.qwen3_moe.modeling_qwen3_moe import (
+    QEffQwen3MoeAttention,
+    QEffQwen3MoeDecoderLayer,
+    QEffQwen3MoeForCausalLM,
+    QEffQwen3MoeModel,
+    QEffQwen3MoeRotaryEmbedding,
+    QEffQwen3MoeSparseMoeBlock,
+)
 from QEfficient.transformers.models.starcoder2.modeling_starcoder2 import (
     QEffStarcoder2Attention,
     QEFFStarcoder2DecoderLayer,
@@ -284,6 +301,7 @@ class CustomOpsTransform(ModuleMappingTransform):
         MllamaTextRMSNorm: CustomRMSNormAIC,
         GraniteRMSNorm: CustomRMSNormAIC,
         GraniteMoeRMSNorm: CustomRMSNormAIC,
+        Qwen3MoeRMSNorm: CustomRMSNormAIC,
     }
 
 
@@ -323,6 +341,13 @@ class KVCacheTransform(ModuleMappingTransform):
         GemmaDecoderLayer: QEffGemmaDecoderLayer,
         GemmaModel: QEffGemmaModel,
         GemmaForCausalLM: QEffGemmaForCausalLM,
+        # Qwen3Moe
+        Qwen3MoeForCausalLM: QEffQwen3MoeForCausalLM,
+        Qwen3MoeModel: QEffQwen3MoeModel,
+        Qwen3MoeDecoderLayer: QEffQwen3MoeDecoderLayer,
+        Qwen3MoeAttention: QEffQwen3MoeAttention,
+        Qwen3MoeRotaryEmbedding: QEffQwen3MoeRotaryEmbedding,
+        Qwen3MoeSparseMoeBlock: QEffQwen3MoeSparseMoeBlock,
         # Gemma2
         Gemma2Attention: QEffGemma2Attention,
         Gemma2DecoderLayer: QEffGemma2DecoderLayer,
diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
new file mode 100644
index 000000000..0b525a3a0
--- /dev/null
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -0,0 +1,599 @@
+from typing import Callable, List, Optional, Tuple, Union
+
+import torch
+import torch.nn.functional as F
+from torch import nn
+from transformers.modeling_outputs import (
+    MoeCausalLMOutputWithPast,
+    MoeModelOutputWithPast,
+)
+from transformers.models.qwen3_moe.modeling_qwen3_moe import (
+    Qwen3MoeAttention,
+    Qwen3MoeConfig,
+    Qwen3MoeDecoderLayer,
+    Qwen3MoeForCausalLM,
+    Qwen3MoeModel,
+    Qwen3MoeRotaryEmbedding,
+    Qwen3MoeSparseMoeBlock,
+    repeat_kv,
+    rotate_half,
+)
+
+from QEfficient.transformers.cache_utils import QEffDynamicCache
+from QEfficient.transformers.modeling_attn_mask_utils import _create_causal_mask
+
+# from QEfficient.transformers.models.llama.modeling_llama import qeff_apply_rotary_pos_emb
+
+
+class QEffQwen3MoeRotaryEmbedding(Qwen3MoeRotaryEmbedding):
+    def __init__(self, config: Qwen3MoeConfig, device=None):
+        super().__init__(config=config)
+
+        # Build here to make `torch.jit.trace` work.
+        self._set_cos_sin_cache(
+            seq_len=self.original_max_seq_len, device=self.inv_freq.device, dtype=torch.get_default_dtype()
+        )
+
+    def _set_cos_sin_cache(self, seq_len, device, dtype):
+        self.max_seq_len_cached = seq_len
+        t = torch.arange(self.max_seq_len_cached, device=device, dtype=torch.int64).type_as(self.inv_freq)
+
+        freqs = torch.outer(t, self.inv_freq)
+
+        emb = torch.cat((freqs, freqs), dim=-1)
+        self.register_buffer("cos_cached", emb.cos().to(dtype), persistent=False)
+        self.register_buffer("sin_cached", emb.sin().to(dtype), persistent=False)
+
+    def forward(self, x, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        if seq_len > self.max_seq_len_cached:
+            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=x.dtype)
+
+        return (
+            self.cos_cached[:seq_len].to(dtype=x.dtype),
+            self.sin_cached[:seq_len].to(dtype=x.dtype),
+        )
+
+
+def qeff_apply_rotary_pos_emb(q, k, cos, sin, position_ids, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`):
+            The position indices of the tokens corresponding to the query and key tensors. For example, this can be
+            used to pass offsetted position ids when working with a KV-cache.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    # breakpoint()
+    cos = cos[position_ids].unsqueeze(unsqueeze_dim)
+    sin = sin[position_ids].unsqueeze(unsqueeze_dim)
+
+    # Apply rotation
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    # Cast back to original dtype
+    return q_embed.to(q.dtype), k_embed.to(k.dtype)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    **kwargs,
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        attn_weights = torch.where(attention_mask, torch.tensor(-10000.0, dtype=torch.float32), attn_weights)
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+class QEffQwen3MoeSparseMoeBlock(Qwen3MoeSparseMoeBlock):
+    def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
+        # # breakpoint()
+        # B, S, D = hidden_states.shape  # [1, 8, 2304]
+        # hidden_states = hidden_states.reshape(-1, D)  # [8, 2304]
+        # T = hidden_states.size(0)  # 8 tokens
+        # router_logits = self.gate(hidden_states)  # [8, 8]
+        # probs = F.softmax(router_logits, dim=-1)  # [8, 8]
+
+        # topk_scores, topk_indices = torch.topk(probs, self.top_k, dim=-1)  # [8, top_k] → topk_k is 2 for Grok1
+        # topk_scores = topk_scores / topk_scores.sum(dim=-1, keepdim=True)  # normalize per-token
+        # topk_scores = topk_scores.to(hidden_states.dtype)  # [8, top_k]
+        # route = torch.zeros((T, self.num_experts), dtype=hidden_states.dtype)
+        # route.scatter_(1, topk_indices, topk_scores)  # [8, num_experts]
+        # final_output = torch.zeros_like(hidden_states)  # [8, 2304]
+
+        # for e, expert in enumerate(self.experts):
+        #     scores = route[:, e].unsqueeze(1)  # [8, 1]
+        #     masked_out = torch.where(
+        #         scores > 0, expert(hidden_states) * scores, 0.0
+        #     )  # # [8, 2304] × [8, 1] → [8, 2304]
+        #     final_output += masked_out  # accumulate expert outputs
+        # return final_output.reshape(B, S, D), router_logits  # ([1, 8, 2304], [8, num_experts])
+
+        B, S, H = hidden_states.shape
+        T = B * S
+        x = hidden_states.view(T, H)
+
+        router_logits = self.gate(x)  # [T, E]
+        prob = F.softmax(router_logits, -1, dtype=torch.float)
+        top_w, top_i = torch.topk(prob, self.top_k, -1)
+        if self.norm_topk_prob:  # only diff with mixtral sparse moe block!
+            top_w /= top_w.sum(-1, keepdim=True)
+        top_w = top_w.to(x.dtype)
+
+        # Create 2 expert idx based on the topk
+        expert1_idx, expert2_idx, expert3_idx, expert4_idx, expert5_idx, expert6_idx, expert7_idx, expert8_idx = (
+            top_i[:, 0],
+            top_i[:, 1],
+            top_i[:, 2],
+            top_i[:, 3],
+            top_i[:, 4],
+            top_i[:, 5],
+            top_i[:, 6],
+            top_i[:, 7],
+        )  # [T]
+        weight1, weight2, weight3, weight4, weight5, weight6, weight7, weight8 = (
+            top_w[:, 0],
+            top_w[:, 1],
+            top_w[:, 2],
+            top_w[:, 3],
+            top_w[:, 4],
+            top_w[:, 5],
+            top_w[:, 6],
+            top_w[:, 7],
+        )  # [T]
+
+        Inter = 768
+        upgate1 = x.new_zeros((T, Inter))
+        upgate2 = x.new_zeros((T, Inter))
+        upgate3 = x.new_zeros((T, Inter))
+        upgate4 = x.new_zeros((T, Inter))
+        upgate5 = x.new_zeros((T, Inter))
+        upgate6 = x.new_zeros((T, Inter))
+        upgate7 = x.new_zeros((T, Inter))
+        upgate8 = x.new_zeros((T, Inter))
+
+        expert_out1 = x.new_zeros((T, H))
+        expert_out2 = x.new_zeros((T, H))
+        expert_out3 = x.new_zeros((T, H))
+        expert_out4 = x.new_zeros((T, H))
+        expert_out5 = x.new_zeros((T, H))
+        expert_out6 = x.new_zeros((T, H))
+        expert_out7 = x.new_zeros((T, H))
+        expert_out8 = x.new_zeros((T, H))
+
+        for e in range(self.num_experts):
+            exp = self.experts[e]
+            mask1 = (expert1_idx == e).unsqueeze(1)  # [T, 1]
+            mask2 = (expert2_idx == e).unsqueeze(1)  # [T, 1]
+            mask3 = (expert3_idx == e).unsqueeze(1)  # [T, 1]
+            mask4 = (expert4_idx == e).unsqueeze(1)  # [T, 1]
+            mask5 = (expert5_idx == e).unsqueeze(1)  # [T, 1]
+            mask6 = (expert6_idx == e).unsqueeze(1)  # [T, 1]
+            mask7 = (expert7_idx == e).unsqueeze(1)  # [T, 1]
+            mask8 = (expert8_idx == e).unsqueeze(1)  # [T, 1]
+
+            # breakpoint()
+            hidden_gate = (exp.act_fn(exp.gate_proj(x))) * exp.up_proj(x)
+            # hidden_gate=exp.down_proj(hidden_gate)
+
+            # Accumulate weighted contributions
+            upgate1 += torch.where(mask1, hidden_gate, torch.zeros_like(upgate1))
+            upgate2 += torch.where(mask2, hidden_gate, torch.zeros_like(upgate2))
+            upgate3 += torch.where(mask3, hidden_gate, torch.zeros_like(upgate3))
+            upgate4 += torch.where(mask4, hidden_gate, torch.zeros_like(upgate4))
+            upgate5 += torch.where(mask5, hidden_gate, torch.zeros_like(upgate5))
+            upgate6 += torch.where(mask6, hidden_gate, torch.zeros_like(upgate6))
+            upgate7 += torch.where(mask7, hidden_gate, torch.zeros_like(upgate7))
+            upgate8 += torch.where(mask8, hidden_gate, torch.zeros_like(upgate8))
+
+        for e in range(self.num_experts):
+            exp = self.experts[e]
+            mask1 = (expert1_idx == e).unsqueeze(1)
+            mask2 = (expert2_idx == e).unsqueeze(1)
+            mask3 = (expert3_idx == e).unsqueeze(1)  # [T, 1]
+            mask4 = (expert4_idx == e).unsqueeze(1)  # [T, 1]
+            mask5 = (expert5_idx == e).unsqueeze(1)  # [T, 1]
+            mask6 = (expert6_idx == e).unsqueeze(1)  # [T, 1]
+            mask7 = (expert7_idx == e).unsqueeze(1)  # [T, 1]
+            mask8 = (expert8_idx == e).unsqueeze(1)
+            # breakpoint()
+            expert_out1 += torch.where(
+                mask1, exp.down_proj(upgate1) * weight1.unsqueeze(1), torch.zeros_like(expert_out1)
+            )
+            expert_out2 += torch.where(
+                mask2, exp.down_proj(upgate2) * weight2.unsqueeze(1), torch.zeros_like(expert_out2)
+            )
+            expert_out3 += torch.where(
+                mask3, exp.down_proj(upgate3) * weight3.unsqueeze(1), torch.zeros_like(expert_out3)
+            )
+            expert_out4 += torch.where(
+                mask4, exp.down_proj(upgate4) * weight4.unsqueeze(1), torch.zeros_like(expert_out4)
+            )
+            expert_out5 += torch.where(
+                mask5, exp.down_proj(upgate5) * weight5.unsqueeze(1), torch.zeros_like(expert_out5)
+            )
+            expert_out6 += torch.where(
+                mask6, exp.down_proj(upgate6) * weight6.unsqueeze(1), torch.zeros_like(expert_out6)
+            )
+            expert_out7 += torch.where(
+                mask7, exp.down_proj(upgate7) * weight7.unsqueeze(1), torch.zeros_like(expert_out7)
+            )
+            expert_out8 += torch.where(
+                mask8, exp.down_proj(upgate8) * weight8.unsqueeze(1), torch.zeros_like(expert_out8)
+            )
+
+        expert_out = (
+            expert_out1
+            + expert_out2
+            + expert_out3
+            + expert_out4
+            + expert_out5
+            + expert_out6
+            + expert_out7
+            + expert_out8
+        )
+        return expert_out.view(B, S, H), router_logits
+
+
+class QEffQwen3MoeAttention(Qwen3MoeAttention):
+    def __qeff_init__(self):
+        self.rotary_emb = QEffQwen3MoeRotaryEmbedding(config=self.config)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor],
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        batch_index: Optional[torch.LongTensor] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_norm(self.q_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
+        key_states = self.k_norm(self.k_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
+        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        # breakpoint()
+        kv_seq_len = key_states.shape[-2]
+        kv_seq_len = past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+
+        # query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+        query_states, key_states = qeff_apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            # sin and cos are specific to RoPE models; cache_position needed for the static cache
+            cache_kwargs = {"sin": sin, "cos": cos, "batch_index": batch_index, "position_ids": position_ids}
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = eager_attention_forward
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights, past_key_value
+
+
+class QEffQwen3MoeDecoderLayer(Qwen3MoeDecoderLayer):
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        output_router_logits: Optional[bool] = False,
+        batch_index: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        # position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,  # necessary, but kept here for BC
+        **kwargs,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, sequence_length)` where padding elements are indicated by 0.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            output_router_logits (`bool`, *optional*):
+                Whether or not to return the logits of all the routers. They are useful for computing the router loss,
+                and should not be returned during inference.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+            cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
+                Indices depicting the position of the input sequence tokens in the sequence.
+            position_embeddings (`Tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*):
+                Tuple containing the cosine and sine positional embeddings of shape `(batch_size, seq_len, head_dim)`,
+                with `head_dim` being the embedding dimension of each attention head.
+            kwargs (`dict`, *optional*):
+                Arbitrary kwargs to be ignored, used for FSDP and other methods that injects code
+                into the model
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            batch_index=batch_index,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+
+        hidden_states = self.mlp(hidden_states)
+        if isinstance(hidden_states, tuple):
+            hidden_states, router_logits = hidden_states
+        else:
+            router_logits = None
+
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if output_router_logits:
+            outputs += (router_logits,)
+
+        return outputs
+
+
+class QEffQwen3MoeModel(Qwen3MoeModel):
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        batch_index: Optional[torch.LongTensor] = None,
+        output_hidden_states: Optional[bool] = None,
+        output_router_logits: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+    ) -> MoeModelOutputWithPast:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_router_logits = (
+            output_router_logits if output_router_logits is not None else self.config.output_router_logits
+        )
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        # if use_cache and past_key_values is None:
+        #     past_key_values = DynamicCache()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        past_key_values_length = 0
+        if past_key_values is not None:
+            past_key_values_length = past_key_values[0][0].shape[2]
+            # seq_length_with_past = seq_length_with_past + past_key_values_length
+
+        past_key_values = QEffDynamicCache.from_legacy_cache(past_key_values)
+
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = _create_causal_mask(position_ids=position_ids, target_length=past_key_values_length)
+
+        hidden_states = inputs_embeds
+        # breakpoint()
+        # create position embeddings to be shared across the decoder layers
+        # position_embeddings = self.rotary_emb(hidden_states, position_ids)
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        all_router_logits = () if output_router_logits else None
+
+        for decoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            layer_outputs = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_values,
+                batch_index=batch_index,
+                output_attentions=output_attentions,
+                output_router_logits=output_router_logits,
+                use_cache=use_cache,
+                cache_position=cache_position,
+                # position_embeddings=position_embeddings,
+            )
+
+            hidden_states = layer_outputs[0]
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+            if output_router_logits:
+                all_router_logits += (layer_outputs[-1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        past_key_values = past_key_values.to_legacy_cache()
+
+        return MoeModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            router_logits=all_router_logits,
+        )
+
+
+class QEffQwen3MoeForCausalLM(Qwen3MoeForCausalLM):
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        batch_index: Optional[torch.LongTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        output_router_logits: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs,
+    ) -> MoeCausalLMOutputWithPast:
+        r"""
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+            logits_to_keep (`int` or `torch.Tensor`, *optional*):
+                If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+                `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+                token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+                If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+                This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, Qwen3MoeForCausalLM
+
+        >>> model = Qwen3MoeForCausalLM.from_pretrained("Qwen/Qwen3-MoE-15B-A2B")
+        >>> tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-MoE-15B-A2B")
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+        # breakpoint()
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_router_logits = (
+            output_router_logits if output_router_logits is not None else self.config.output_router_logits
+        )
+
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        # breakpoint()
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs: MoeModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            batch_index=batch_index,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            output_router_logits=output_router_logits,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        # slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        # logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        # loss = None
+        # if labels is not None:
+        #     loss = self.loss_function(logits, labels, self.vocab_size, **kwargs)
+
+        # aux_loss = None
+        # if output_router_logits:
+        #     aux_loss = load_balancing_loss_func(
+        #         outputs.router_logits,
+        #         self.num_experts,
+        #         self.num_experts_per_tok,
+        #         attention_mask,
+        #     )
+        #     if labels is not None:
+        #         loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
+        # breakpoint()
+        logit_idx = position_ids.to(torch.int32).argmax(1, keepdim=True)
+        hidden_states = outputs[0][torch.arange(position_ids.shape[0]).view(-1, 1), logit_idx]
+        logits = self.lm_head(hidden_states)
+        # logits = logits * self.output_multiplier_scale
+        logits = logits.float()
+
+        return MoeCausalLMOutputWithPast(
+            # loss=loss,
+            # aux_loss=aux_loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            router_logits=outputs.router_logits,
+        )

From a497e33a10a19ca2a94b8379a03d30de2bf73997 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Fri, 16 May 2025 09:26:18 +0000
Subject: [PATCH 02/13] Dynamic code for Moe Block

Signed-off-by: Dipankar Sarkar <quic_dipankar@quicinc.com>
---
 .../models/qwen3_moe/modeling_qwen3_moe.py    | 155 +++---------------
 1 file changed, 27 insertions(+), 128 deletions(-)

diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index 0b525a3a0..873639c77 100644
--- a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -112,28 +112,6 @@ def eager_attention_forward(
 
 class QEffQwen3MoeSparseMoeBlock(Qwen3MoeSparseMoeBlock):
     def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
-        # # breakpoint()
-        # B, S, D = hidden_states.shape  # [1, 8, 2304]
-        # hidden_states = hidden_states.reshape(-1, D)  # [8, 2304]
-        # T = hidden_states.size(0)  # 8 tokens
-        # router_logits = self.gate(hidden_states)  # [8, 8]
-        # probs = F.softmax(router_logits, dim=-1)  # [8, 8]
-
-        # topk_scores, topk_indices = torch.topk(probs, self.top_k, dim=-1)  # [8, top_k] → topk_k is 2 for Grok1
-        # topk_scores = topk_scores / topk_scores.sum(dim=-1, keepdim=True)  # normalize per-token
-        # topk_scores = topk_scores.to(hidden_states.dtype)  # [8, top_k]
-        # route = torch.zeros((T, self.num_experts), dtype=hidden_states.dtype)
-        # route.scatter_(1, topk_indices, topk_scores)  # [8, num_experts]
-        # final_output = torch.zeros_like(hidden_states)  # [8, 2304]
-
-        # for e, expert in enumerate(self.experts):
-        #     scores = route[:, e].unsqueeze(1)  # [8, 1]
-        #     masked_out = torch.where(
-        #         scores > 0, expert(hidden_states) * scores, 0.0
-        #     )  # # [8, 2304] × [8, 1] → [8, 2304]
-        #     final_output += masked_out  # accumulate expert outputs
-        # return final_output.reshape(B, S, D), router_logits  # ([1, 8, 2304], [8, num_experts])
-
         B, S, H = hidden_states.shape
         T = B * S
         x = hidden_states.view(T, H)
@@ -145,119 +123,40 @@ def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tens
             top_w /= top_w.sum(-1, keepdim=True)
         top_w = top_w.to(x.dtype)
 
-        # Create 2 expert idx based on the topk
-        expert1_idx, expert2_idx, expert3_idx, expert4_idx, expert5_idx, expert6_idx, expert7_idx, expert8_idx = (
-            top_i[:, 0],
-            top_i[:, 1],
-            top_i[:, 2],
-            top_i[:, 3],
-            top_i[:, 4],
-            top_i[:, 5],
-            top_i[:, 6],
-            top_i[:, 7],
-        )  # [T]
-        weight1, weight2, weight3, weight4, weight5, weight6, weight7, weight8 = (
-            top_w[:, 0],
-            top_w[:, 1],
-            top_w[:, 2],
-            top_w[:, 3],
-            top_w[:, 4],
-            top_w[:, 5],
-            top_w[:, 6],
-            top_w[:, 7],
-        )  # [T]
-
-        Inter = 768
-        upgate1 = x.new_zeros((T, Inter))
-        upgate2 = x.new_zeros((T, Inter))
-        upgate3 = x.new_zeros((T, Inter))
-        upgate4 = x.new_zeros((T, Inter))
-        upgate5 = x.new_zeros((T, Inter))
-        upgate6 = x.new_zeros((T, Inter))
-        upgate7 = x.new_zeros((T, Inter))
-        upgate8 = x.new_zeros((T, Inter))
-
-        expert_out1 = x.new_zeros((T, H))
-        expert_out2 = x.new_zeros((T, H))
-        expert_out3 = x.new_zeros((T, H))
-        expert_out4 = x.new_zeros((T, H))
-        expert_out5 = x.new_zeros((T, H))
-        expert_out6 = x.new_zeros((T, H))
-        expert_out7 = x.new_zeros((T, H))
-        expert_out8 = x.new_zeros((T, H))
+        expert_idx = []
+        weights = []
+        for i in range(self.top_k):
+            expert_idx.append(top_i[:, i])
+            weights.append(top_w[:, i])
+
+        # I = self.config.ffn_dim
+        Inter = 768  # TODO: Find a way to identify from config # Intermediate Size
+        upgate = []
+        expert_out = []
+        for i in range(self.top_k):
+            upgate.append(x.new_zeros((T, Inter)))
+            expert_out.append(x.new_zeros((T, H)))
 
         for e in range(self.num_experts):
             exp = self.experts[e]
-            mask1 = (expert1_idx == e).unsqueeze(1)  # [T, 1]
-            mask2 = (expert2_idx == e).unsqueeze(1)  # [T, 1]
-            mask3 = (expert3_idx == e).unsqueeze(1)  # [T, 1]
-            mask4 = (expert4_idx == e).unsqueeze(1)  # [T, 1]
-            mask5 = (expert5_idx == e).unsqueeze(1)  # [T, 1]
-            mask6 = (expert6_idx == e).unsqueeze(1)  # [T, 1]
-            mask7 = (expert7_idx == e).unsqueeze(1)  # [T, 1]
-            mask8 = (expert8_idx == e).unsqueeze(1)  # [T, 1]
-
-            # breakpoint()
+            mask = []
+            for i in range(self.top_k):
+                mask.append((expert_idx[i] == e).unsqueeze(1))
             hidden_gate = (exp.act_fn(exp.gate_proj(x))) * exp.up_proj(x)
-            # hidden_gate=exp.down_proj(hidden_gate)
-
-            # Accumulate weighted contributions
-            upgate1 += torch.where(mask1, hidden_gate, torch.zeros_like(upgate1))
-            upgate2 += torch.where(mask2, hidden_gate, torch.zeros_like(upgate2))
-            upgate3 += torch.where(mask3, hidden_gate, torch.zeros_like(upgate3))
-            upgate4 += torch.where(mask4, hidden_gate, torch.zeros_like(upgate4))
-            upgate5 += torch.where(mask5, hidden_gate, torch.zeros_like(upgate5))
-            upgate6 += torch.where(mask6, hidden_gate, torch.zeros_like(upgate6))
-            upgate7 += torch.where(mask7, hidden_gate, torch.zeros_like(upgate7))
-            upgate8 += torch.where(mask8, hidden_gate, torch.zeros_like(upgate8))
+            for i in range(self.top_k):
+                upgate[i] += torch.where(mask[i], hidden_gate, torch.zeros_like(upgate[i]))
 
         for e in range(self.num_experts):
             exp = self.experts[e]
-            mask1 = (expert1_idx == e).unsqueeze(1)
-            mask2 = (expert2_idx == e).unsqueeze(1)
-            mask3 = (expert3_idx == e).unsqueeze(1)  # [T, 1]
-            mask4 = (expert4_idx == e).unsqueeze(1)  # [T, 1]
-            mask5 = (expert5_idx == e).unsqueeze(1)  # [T, 1]
-            mask6 = (expert6_idx == e).unsqueeze(1)  # [T, 1]
-            mask7 = (expert7_idx == e).unsqueeze(1)  # [T, 1]
-            mask8 = (expert8_idx == e).unsqueeze(1)
-            # breakpoint()
-            expert_out1 += torch.where(
-                mask1, exp.down_proj(upgate1) * weight1.unsqueeze(1), torch.zeros_like(expert_out1)
-            )
-            expert_out2 += torch.where(
-                mask2, exp.down_proj(upgate2) * weight2.unsqueeze(1), torch.zeros_like(expert_out2)
-            )
-            expert_out3 += torch.where(
-                mask3, exp.down_proj(upgate3) * weight3.unsqueeze(1), torch.zeros_like(expert_out3)
-            )
-            expert_out4 += torch.where(
-                mask4, exp.down_proj(upgate4) * weight4.unsqueeze(1), torch.zeros_like(expert_out4)
-            )
-            expert_out5 += torch.where(
-                mask5, exp.down_proj(upgate5) * weight5.unsqueeze(1), torch.zeros_like(expert_out5)
-            )
-            expert_out6 += torch.where(
-                mask6, exp.down_proj(upgate6) * weight6.unsqueeze(1), torch.zeros_like(expert_out6)
-            )
-            expert_out7 += torch.where(
-                mask7, exp.down_proj(upgate7) * weight7.unsqueeze(1), torch.zeros_like(expert_out7)
-            )
-            expert_out8 += torch.where(
-                mask8, exp.down_proj(upgate8) * weight8.unsqueeze(1), torch.zeros_like(expert_out8)
-            )
-
-        expert_out = (
-            expert_out1
-            + expert_out2
-            + expert_out3
-            + expert_out4
-            + expert_out5
-            + expert_out6
-            + expert_out7
-            + expert_out8
-        )
-        return expert_out.view(B, S, H), router_logits
+            mask = []
+            for i in range(self.top_k):
+                mask.append((expert_idx[i] == e).unsqueeze(1))
+                expert_out[i] += torch.where(
+                    mask[i], exp.down_proj(upgate[i]) * (weights[i].unsqueeze(1)), torch.zeros_like(expert_out[i])
+                )
+
+        expert_out_sum = sum(expert_out)
+        return expert_out_sum.view(B, S, H), router_logits
 
 
 class QEffQwen3MoeAttention(Qwen3MoeAttention):

From 65b0eabbd8eda11985f486326215c0081cc3d30e Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Fri, 16 May 2025 09:37:56 +0000
Subject: [PATCH 03/13] Fix for Repopointer issue

Signed-off-by: Dipankar Sarkar <quic_dipankar@quicinc.com>
---
 QEfficient/transformers/models/qwen3_moe/__init__.py       | 6 ++++++
 .../transformers/models/qwen3_moe/modeling_qwen3_moe.py    | 7 +++++++
 2 files changed, 13 insertions(+)
 create mode 100644 QEfficient/transformers/models/qwen3_moe/__init__.py

diff --git a/QEfficient/transformers/models/qwen3_moe/__init__.py b/QEfficient/transformers/models/qwen3_moe/__init__.py
new file mode 100644
index 000000000..72ba36c8a
--- /dev/null
+++ b/QEfficient/transformers/models/qwen3_moe/__init__.py
@@ -0,0 +1,6 @@
+# -----------------------------------------------------------------------------
+#
+# Copyright (c) 2025 Qualcomm Innovation Center, Inc. All rights reserved.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+# -----------------------------------------------------------------------------
diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index 873639c77..622ec6ca2 100644
--- a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -1,3 +1,10 @@
+# -----------------------------------------------------------------------------
+#
+# Copyright (c) 2025 Qualcomm Innovation Center, Inc. All rights reserved.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+# -----------------------------------------------------------------------------
+
 from typing import Callable, List, Optional, Tuple, Union
 
 import torch

From 03df777427a524099e2049a2afac83eddcc9e19c Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Wed, 21 May 2025 17:40:32 +0000
Subject: [PATCH 04/13] upgrading transformer to 4.51.3

Signed-off-by: Dipankar Sarkar <quic_dipankar@quicinc.com>
---
 pyproject.toml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pyproject.toml b/pyproject.toml
index 648d2ce4e..d718bb6c9 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -19,7 +19,7 @@ classifiers = [
 ]
 requires-python = ">=3.8,<3.11"
 dependencies = [
-    "transformers==4.50.0",
+    "transformers==4.51.3",
     "huggingface-hub==0.27.0",
     "hf_transfer==0.1.9",
     "peft==0.13.2",

From 1ba817049f913a08eea0e7bc61ce48ae1d172e6b Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Wed, 21 May 2025 18:40:33 +0000
Subject: [PATCH 05/13] upgrading transformer to 4.51.0

Signed-off-by: Dipankar Sarkar <quic_dipankar@quicinc.com>
---
 pyproject.toml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pyproject.toml b/pyproject.toml
index d718bb6c9..aefb29202 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -19,7 +19,7 @@ classifiers = [
 ]
 requires-python = ">=3.8,<3.11"
 dependencies = [
-    "transformers==4.51.3",
+    "transformers==4.51.0",
     "huggingface-hub==0.27.0",
     "hf_transfer==0.1.9",
     "peft==0.13.2",

From a82861b3c69b8733b81541af861cbf7aa37400e7 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <dipankar@qti.qualcomm.com>
Date: Tue, 1 Jul 2025 05:24:41 +0000
Subject: [PATCH 06/13] Cleanup done for qwen3moe

Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
---
 .../models/qwen3_moe/modeling_qwen3_moe.py    | 36 +++----------------
 1 file changed, 4 insertions(+), 32 deletions(-)

diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index 622ec6ca2..a8904fc5d 100644
--- a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -29,8 +29,6 @@
 from QEfficient.transformers.cache_utils import QEffDynamicCache
 from QEfficient.transformers.modeling_attn_mask_utils import _create_causal_mask
 
-# from QEfficient.transformers.models.llama.modeling_llama import qeff_apply_rotary_pos_emb
-
 
 class QEffQwen3MoeRotaryEmbedding(Qwen3MoeRotaryEmbedding):
     def __init__(self, config: Qwen3MoeConfig, device=None):
@@ -137,7 +135,8 @@ def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tens
             weights.append(top_w[:, i])
 
         # I = self.config.ffn_dim
-        Inter = 768  # TODO: Find a way to identify from config # Intermediate Size
+        Inter = self.config.ffn_dim  # TODO: Find a way to identify from config # Intermediate Size
+        breakpoint()
         upgate = []
         expert_out = []
         for i in range(self.top_k):
@@ -323,16 +322,12 @@ def forward(
         if (input_ids is None) ^ (inputs_embeds is not None):
             raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
 
-        # if use_cache and past_key_values is None:
-        #     past_key_values = DynamicCache()
-
         if inputs_embeds is None:
             inputs_embeds = self.embed_tokens(input_ids)
 
         past_key_values_length = 0
         if past_key_values is not None:
             past_key_values_length = past_key_values[0][0].shape[2]
-            # seq_length_with_past = seq_length_with_past + past_key_values_length
 
         past_key_values = QEffDynamicCache.from_legacy_cache(past_key_values)
 
@@ -342,9 +337,6 @@ def forward(
         causal_mask = _create_causal_mask(position_ids=position_ids, target_length=past_key_values_length)
 
         hidden_states = inputs_embeds
-        # breakpoint()
-        # create position embeddings to be shared across the decoder layers
-        # position_embeddings = self.rotary_emb(hidden_states, position_ids)
 
         # decoder layers
         all_hidden_states = () if output_hidden_states else None
@@ -365,7 +357,6 @@ def forward(
                 output_router_logits=output_router_logits,
                 use_cache=use_cache,
                 cache_position=cache_position,
-                # position_embeddings=position_embeddings,
             )
 
             hidden_states = layer_outputs[0]
@@ -442,7 +433,7 @@ def forward(
         >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
         "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
         ```"""
-        # breakpoint()
+
         output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
         output_router_logits = (
             output_router_logits if output_router_logits is not None else self.config.output_router_logits
@@ -451,7 +442,7 @@ def forward(
         output_hidden_states = (
             output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
         )
-        # breakpoint()
+
         # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
         outputs: MoeModelOutputWithPast = self.model(
             input_ids=input_ids,
@@ -469,25 +460,6 @@ def forward(
         )
 
         hidden_states = outputs.last_hidden_state
-        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
-        # slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
-        # logits = self.lm_head(hidden_states[:, slice_indices, :])
-
-        # loss = None
-        # if labels is not None:
-        #     loss = self.loss_function(logits, labels, self.vocab_size, **kwargs)
-
-        # aux_loss = None
-        # if output_router_logits:
-        #     aux_loss = load_balancing_loss_func(
-        #         outputs.router_logits,
-        #         self.num_experts,
-        #         self.num_experts_per_tok,
-        #         attention_mask,
-        #     )
-        #     if labels is not None:
-        #         loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
-        # breakpoint()
         logit_idx = position_ids.to(torch.int32).argmax(1, keepdim=True)
         hidden_states = outputs[0][torch.arange(position_ids.shape[0]).view(-1, 1), logit_idx]
         logits = self.lm_head(hidden_states)

From 3a57b236a5ee58cface07d8179bde3d128d60205 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Mon, 4 Aug 2025 07:48:30 +0000
Subject: [PATCH 07/13] Adding Qwen3Moe to test

Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
---
 tests/transformers/models/test_causal_lm_models.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tests/transformers/models/test_causal_lm_models.py b/tests/transformers/models/test_causal_lm_models.py
index 3195c4828..3aabc401a 100644
--- a/tests/transformers/models/test_causal_lm_models.py
+++ b/tests/transformers/models/test_causal_lm_models.py
@@ -31,6 +31,7 @@
     "tiiuae/falcon-7b",
     "Qwen/Qwen2-0.5B",
     "bigcode/starcoder2-3b",
+    "Qwen/Qwen3-30B-A3B",
     "Felladrin/Minueza-32M-Base",
     "wtang06/mpt-125m-c4",
     "hakurei/gpt-j-random-tinier",

From 01eaa397d4971e1d187112d130af263695e07f27 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <dipankar@qti.qualcomm.com>
Date: Tue, 5 Aug 2025 17:53:33 +0530
Subject: [PATCH 08/13] Update modeling_qwen3_moe.py

Minor Changes

Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
---
 QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index a8904fc5d..bd84abb8a 100644
--- a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -135,7 +135,7 @@ def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tens
             weights.append(top_w[:, i])
 
         # I = self.config.ffn_dim
-        Inter = self.config.ffn_dim  # TODO: Find a way to identify from config # Intermediate Size
+        Inter = 768
         breakpoint()
         upgate = []
         expert_out = []

From 5bc95c7170e0fda55926b24db9aa36f495b99459 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Tue, 12 Aug 2025 07:58:11 +0000
Subject: [PATCH 09/13] Added qwen3moe to custom config

Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
---
 .../models/custom_tiny_model_configs.json      | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/tests/transformers/models/custom_tiny_model_configs.json b/tests/transformers/models/custom_tiny_model_configs.json
index 2da2c95a3..8fc315862 100644
--- a/tests/transformers/models/custom_tiny_model_configs.json
+++ b/tests/transformers/models/custom_tiny_model_configs.json
@@ -39,6 +39,7 @@
       "rotary_dim": 16
     }
   },
+  
   {
     "model_name": "microsoft/Phi-3-mini-4k-instruct",
     "model_type": "phi3",
@@ -65,6 +66,23 @@
       "num_key_value_heads": 1
     }
   },
+  {
+    "model_name": "Qwen/Qwen3-30B-A3B",
+    "model_type": "qwen3_moe",
+    "additional_params": {
+      "hidden_size": 256,
+      "intermediate_size": 256,
+      "max_position_embeddings": 128,
+      "max_window_layers": 48,
+      "moe_intermediate_size": 768,
+      "num_attention_heads": 2,
+      "num_experts": 4,
+      "num_experts_per_tok": 2,
+      "num_hidden_layers": 1,
+      "num_key_value_heads": 1,
+      "vocab_size": 151936
+    }
+  },
   {
     "model_name": "Qwen/Qwen2-0.5B",
     "model_type": "qwen2",

From 2e3090ccc54be5e29abf7e714822a67c5a329119 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <dipankar@qti.qualcomm.com>
Date: Tue, 12 Aug 2025 15:22:36 +0530
Subject: [PATCH 10/13] final cleaning

Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
---
 .../transformers/models/qwen3_moe/modeling_qwen3_moe.py     | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index bd84abb8a..c025031ff 100644
--- a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -81,7 +81,7 @@ def qeff_apply_rotary_pos_emb(q, k, cos, sin, position_ids, unsqueeze_dim=1):
     Returns:
         `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
     """
-    # breakpoint()
+    
     cos = cos[position_ids].unsqueeze(unsqueeze_dim)
     sin = sin[position_ids].unsqueeze(unsqueeze_dim)
 
@@ -133,10 +133,7 @@ def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tens
         for i in range(self.top_k):
             expert_idx.append(top_i[:, i])
             weights.append(top_w[:, i])
-
-        # I = self.config.ffn_dim
         Inter = 768
-        breakpoint()
         upgate = []
         expert_out = []
         for i in range(self.top_k):
@@ -185,7 +182,6 @@ def forward(
         query_states = self.q_norm(self.q_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
         key_states = self.k_norm(self.k_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
         value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
-        # breakpoint()
         kv_seq_len = key_states.shape[-2]
         kv_seq_len = past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
         cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)

From 1329a096a04961c738f4ba2f3b4ea5eeda68212f Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Tue, 12 Aug 2025 09:55:54 +0000
Subject: [PATCH 11/13] Final Cleaning for qwenmoe

Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
---
 QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index c025031ff..c129f8740 100644
--- a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -81,7 +81,7 @@ def qeff_apply_rotary_pos_emb(q, k, cos, sin, position_ids, unsqueeze_dim=1):
     Returns:
         `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
     """
-    
+
     cos = cos[position_ids].unsqueeze(unsqueeze_dim)
     sin = sin[position_ids].unsqueeze(unsqueeze_dim)
 
@@ -102,6 +102,7 @@ def eager_attention_forward(
     **kwargs,
 ):
     key_states = repeat_kv(key, module.num_key_value_groups)
+
     value_states = repeat_kv(value, module.num_key_value_groups)
 
     attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling

From 7acb7f36a94c7993a7e55cde3b56e8347dd34260 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Sat, 23 Aug 2025 07:14:10 +0000
Subject: [PATCH 12/13] Comments addresed

Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
---
 .../transformers/models/qwen3_moe/modeling_qwen3_moe.py  | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index c129f8740..061b7e8c9 100644
--- a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -28,6 +28,7 @@
 
 from QEfficient.transformers.cache_utils import QEffDynamicCache
 from QEfficient.transformers.modeling_attn_mask_utils import _create_causal_mask
+from QEfficient.utils.constants import MIN_MASKED_ATTENTION_VALUE
 
 
 class QEffQwen3MoeRotaryEmbedding(Qwen3MoeRotaryEmbedding):
@@ -107,7 +108,9 @@ def eager_attention_forward(
 
     attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
     if attention_mask is not None:
-        attn_weights = torch.where(attention_mask, torch.tensor(-10000.0, dtype=torch.float32), attn_weights)
+        attn_weights = torch.where(
+            attention_mask, torch.tensor(MIN_MASKED_ATTENTION_VALUE, dtype=torch.float32), attn_weights
+        )
 
     attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
     attn_output = torch.matmul(attn_weights, value_states)
@@ -134,6 +137,7 @@ def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tens
         for i in range(self.top_k):
             expert_idx.append(top_i[:, i])
             weights.append(top_w[:, i])
+        breakpoint()
         Inter = 768
         upgate = []
         expert_out = []
@@ -460,12 +464,9 @@ def forward(
         logit_idx = position_ids.to(torch.int32).argmax(1, keepdim=True)
         hidden_states = outputs[0][torch.arange(position_ids.shape[0]).view(-1, 1), logit_idx]
         logits = self.lm_head(hidden_states)
-        # logits = logits * self.output_multiplier_scale
         logits = logits.float()
 
         return MoeCausalLMOutputWithPast(
-            # loss=loss,
-            # aux_loss=aux_loss,
             logits=logits,
             past_key_values=outputs.past_key_values,
             hidden_states=outputs.hidden_states,

From c889b40eb16e480716d6a025c3c4c1b8d171e8d4 Mon Sep 17 00:00:00 2001
From: Dipankar Sarkar <quic_dipankar@quicinc.com>
Date: Sat, 23 Aug 2025 07:15:02 +0000
Subject: [PATCH 13/13] Comments addresed

Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
---
 QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index 061b7e8c9..60af9cb64 100644
--- a/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/QEfficient/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -137,7 +137,6 @@ def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tens
         for i in range(self.top_k):
             expert_idx.append(top_i[:, i])
             weights.append(top_w[:, i])
-        breakpoint()
         Inter = 768
         upgate = []
         expert_out = []