support and optimize janus pro (#12813)

Author: MeouSker77

python/llm/src/ipex_llm/transformers/convert.py

@@ -1066,7 +1066,7 @@ def _optimize_pre(model, qtype=None):
         from ipex_llm.transformers.models.baichuan_m1 import pre_register_inv_freq
         model.apply(pre_register_inv_freq)
     elif model.config.model_type == "multi_modality":
-        pass
+        _optimize_pre(model.language_model)
 
     return model
 
@@ -2012,8 +2012,10 @@ def _optimize_post(model):
         # vision
         vpm_modeling_module_name = model.vision_model.vision_tower.__class__.__module__
         vpm_module = importlib.import_module(vpm_modeling_module_name)
-
         from ipex_llm.transformers.models.janus import vision_attention_forward
         convert_forward(model.vision_model, vpm_module.Attention, vision_attention_forward)
 
+        # llm
+        _optimize_post(model.language_model)
+
     return model

python/llm/src/ipex_llm/transformers/models/janus.py

@@ -0,0 +1,49 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This file is adapted from
+# https://github.com/deepseek-ai/Janus/blob/main/janus/models/siglip_vit.py
+
+import torch
+
+from ipex_llm.transformers.models.common import scaled_dot_product_attention
+
+
+def vision_attention_forward(self, x: torch.Tensor) -> torch.Tensor:
+    B, N, C = x.shape
+    qkv = (
+        self.qkv(x)
+        .reshape(B, N, 3, self.num_heads, self.head_dim)
+        .permute(2, 0, 3, 1, 4)
+    )
+    q, k, v = qkv.unbind(0)
+    q, k = self.q_norm(q), self.k_norm(k)
+
+    if self.fused_attn:
+        # ipex-llm opt: sdpa
+        x = scaled_dot_product_attention(
+            q, k.contiguous(), v.contiguous(), None, False
+        )
+    else:
+        q = q * self.scale
+        attn = q @ k.transpose(-2, -1)
+        attn = attn.softmax(dim=-1)
+        attn = self.attn_drop(attn)
+        x = attn @ v
+
+    x = x.transpose(1, 2).reshape(B, N, C)
+    x = self.proj(x)
+    x = self.proj_drop(x)
+    return x

python/llm/src/ipex_llm/transformers/models/utils.py

@@ -86,7 +86,7 @@ def use_quantize_kv_cache(linear: torch.nn.Module, x: torch.Tensor,
         return os.environ["IPEX_LLM_QUANTIZE_KV_CACHE"] == "1"
     elif os.environ.get("IPEX_LLM_LOW_MEM", None) is not None:
         return os.environ["IPEX_LLM_LOW_MEM"] == "1"
-    elif linear.qtype in [ggml_tensor_qtype["fp16"], ggml_tensor_qtype["bf16"]]:
+    elif linear.weight.dtype != torch.uint8:    # unquantized
         return False
     else:
         device_name = get_xpu_device_name(x.device)