onnx-diagnostic 0.8.0__py3-none-any.whl → 0.8.2__py3-none-any.whl

onnx_diagnostic/torch_models/hghub/hub_data_cached_configs.py

@@ -4903,3 +4903,56 @@ def _ccached_hf_internal_testing_tiny_random_gemma3_for_causal_lm():
             "vocab_size": 262144,
         }
     )
+
+
+def _ccached_qwen_qwen2_5_vl_7b_instruct():
+    "Qwen/Qwen2.5-VL-7B-Instruct"
+    return transformers.Qwen2_5_VLConfig(
+        **{
+            "architectures": ["Qwen2_5_VLForConditionalGeneration"],
+            "attention_dropout": 0.0,
+            "bos_token_id": 151643,
+            "eos_token_id": 151645,
+            "vision_start_token_id": 151652,
+            "vision_end_token_id": 151653,
+            "vision_token_id": 151654,
+            "image_token_id": 151655,
+            "video_token_id": 151656,
+            "hidden_act": "silu",
+            "hidden_size": 3584,
+            "initializer_range": 0.02,
+            "intermediate_size": 18944,
+            "max_position_embeddings": 128000,
+            "max_window_layers": 28,
+            "model_type": "qwen2_5_vl",
+            "num_attention_heads": 28,
+            "num_hidden_layers": 28,
+            "num_key_value_heads": 4,
+            "rms_norm_eps": 1e-06,
+            "rope_theta": 1000000.0,
+            "sliding_window": 32768,
+            "tie_word_embeddings": false,
+            "torch_dtype": "bfloat16",
+            "transformers_version": "4.41.2",
+            "use_cache": true,
+            "use_sliding_window": false,
+            "vision_config": {
+                "depth": 32,
+                "hidden_act": "silu",
+                "hidden_size": 1280,
+                "intermediate_size": 3420,
+                "num_heads": 16,
+                "in_chans": 3,
+                "out_hidden_size": 3584,
+                "patch_size": 14,
+                "spatial_merge_size": 2,
+                "spatial_patch_size": 14,
+                "window_size": 112,
+                "fullatt_block_indexes": [7, 15, 23, 31],
+                "tokens_per_second": 2,
+                "temporal_patch_size": 2,
+            },
+            "rope_scaling": {"type": "mrope", "mrope_section": [16, 24, 24]},
+            "vocab_size": 152064,
+        }
+    )

onnx_diagnostic/torch_models/hghub/model_inputs.py

@@ -3,7 +3,7 @@ import inspect
 import os
 import pprint
 import time
-from typing import Any, Dict, Optional, Tuple
+from typing import Any, Callable, Dict, Optional, Tuple
 import torch
 import transformers
 from ...helpers.config_helper import update_config, build_diff_config
@@ -53,6 +53,7 @@ def get_untrained_model_with_inputs(
     add_second_input: int = 1,
     subfolder: Optional[str] = None,
     use_only_preinstalled: bool = False,
+    config_reduction: Optional[Callable[[Any, str], Dict]] = None,
 ) -> Dict[str, Any]:
     """
     Gets a non initialized model similar to the original model
@@ -75,6 +76,12 @@ def get_untrained_model_with_inputs(
         supports different shapes
     :param subfolder: subfolder to use for this model id
     :param use_only_preinstalled: use only preinstalled version
+    :param config_reduction: if specified, this function is used to reduce the
+        model size by tweaking the configuration, it returns a dictionary with values
+        to update, if empty, function:`reduce_model_config
+        <onnx_diagnostic.torch_models.hghub.reduce_model_config>`,
+        this function takes a configuration and a task (string)
+        as arguments
     :return: dictionary with a model, inputs, dynamic shapes, and the configuration,
         some necessary rewriting as well
 
@@ -157,7 +164,13 @@ def get_untrained_model_with_inputs(
 
         # updating the configuration
         config0 = copy.deepcopy(config)
-        mkwargs = reduce_model_config(config, task) if not same_as_pretrained else {}
+        if config_reduction:
+            assert (
+                not same_as_pretrained
+            ), "config_reduction should be None if same_as_pretrained is True"
+            mkwargs = config_reduction(config, task)
+        else:
+            mkwargs = reduce_model_config(config, task) if not same_as_pretrained else {}
         if model_kwargs:
             for k, v in model_kwargs.items():
                 if isinstance(v, dict):