initial snapshot

README.md

@@ -0,0 +1,31 @@
+---
+license: apache-2.0
+tags:
+  - test
+  - fixture
+---
+
+# nano-gv
+
+Random-init ~16 MB CI fixture for the [sendnn-inference](https://github.com/torch-spyre/sendnn-inference)
+plugin's async multimodal-encoder tests. Generated by `tools/build_nano_gv.py`
+from [`ibm-granite/granite-vision-3.2-2b`](https://huggingface.co/ibm-granite/granite-vision-3.2-2b)'s config.
+
+**Not a functional model — outputs are meaningless.** Do not use for
+inference. The architecture identifiers (`LlavaNextForConditionalGeneration`
++ granite text + siglip vision) match the source, but every dim is
+shrunk to the minimum that still exercises the plugin's MM code paths:
+
+| component | source           | nano                |
+|-----------|------------------|---------------------|
+| text hidden      | 2048       | 128                 |
+| text layers      | 40         | 2                   |
+| text heads       | 32         | 2                   |
+| text head_dim    | 64         | 128 (plugin req)    |
+| vision hidden    | 1152       | 64                  |
+| vision layers    | 27         | 4                   |
+| image size       | 384        | 112                 |
+
+The tokenizer + processor are copied from the source verbatim (with
+image-size fields rescaled to 112) so `image_token_index` and the chat
+template stay valid.

chat_template.jinja

@@ -0,0 +1,40 @@
+{%- if tools %}
+    {{- '<|start_of_role|>available_tools<|end_of_role|>
+' }}
+    {%- for tool in tools %}
+    {{- tool | tojson(indent=4) }}
+    {%- if not loop.last %}
+        {{- '
+
+' }}
+    {%- endif %}
+    {%- endfor %}
+    {{- '<|end_of_text|>
+' }}
+{%- endif %}
+{%- for message in messages if message['role'] == 'system'%}{% else %}<|system|>
+A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
+{% endfor %}{%- for message in messages %}
+    {%- if message['role'] == 'system' %}
+    {{- '<|system|>
+' + message['content'][0]['text'] + '
+' }}
+    {%- elif message['role'] == 'user' %}<|user|>
+ {# Render all images first #}{% for content in message['content'] | selectattr('type', 'equalto', 'image') %}{{ '<image>
+' }}{% endfor %}{# Render all text next #}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{{ content['text'] + '
+' }}{% endfor %}
+{%- elif message['role'] == 'assistant' %}
+    {{- '<|assistant|>
+'  + message['content'][0]['text']  + '<|end_of_text|>' }}
+    {%- elif message['role'] == 'assistant_tool_call' %}
+    {{- '<|start_of_role|>assistant<|end_of_role|><|tool_call|>' + message['content'][0]['text']  + '<|end_of_text|>
+' }}
+    {%- elif message['role'] == 'tool_response' %}
+    {{- '<|start_of_role|>tool_response<|end_of_role|>' + message['content'][0]['text'] + '<|end_of_text|>
+' }}
+    {%- endif %}
+    {%- if loop.last and add_generation_prompt %}
+    {{- '<|assistant|>
+' }}
+    {%- endif %}
+{%- endfor %}

config.json

@@ -0,0 +1,82 @@
+{
+  "architectures": [
+    "LlavaNextForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "image_grid_pinpoints": [
+    [
+      112,
+      112
+    ],
+    [
+      112,
+      224
+    ],
+    [
+      224,
+      112
+    ]
+  ],
+  "image_seq_length": 576,
+  "image_token_index": 49155,
+  "model_type": "llava_next",
+  "multimodal_projector_bias": true,
+  "projector_hidden_act": "gelu",
+  "text_config": {
+    "architectures": [
+      "GraniteForCausalLM"
+    ],
+    "attention_bias": false,
+    "attention_dropout": 0.1,
+    "attention_multiplier": 0.015625,
+    "bos_token_id": 0,
+    "dtype": "bfloat16",
+    "embedding_multiplier": 12.0,
+    "eos_token_id": 0,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 128,
+    "initializer_range": 0.02,
+    "intermediate_size": 128,
+    "logits_scaling": 8.0,
+    "max_position_embeddings": 131072,
+    "mlp_bias": false,
+    "model_type": "granite",
+    "num_attention_heads": 2,
+    "num_hidden_layers": 2,
+    "num_key_value_heads": 2,
+    "pad_token_id": 0,
+    "residual_multiplier": 0.22,
+    "rms_norm_eps": 1e-05,
+    "rope_parameters": {
+      "rope_theta": 300000,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": true,
+    "use_cache": true,
+    "vocab_size": 49156
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.8.1",
+  "use_image_newline_parameter": true,
+  "vision_config": {
+    "attention_dropout": 0.0,
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 64,
+    "image_size": 112,
+    "intermediate_size": 128,
+    "layer_norm_eps": 1e-06,
+    "model_type": "siglip_vision_model",
+    "num_attention_heads": 4,
+    "num_channels": 3,
+    "num_hidden_layers": 4,
+    "patch_size": 14
+  },
+  "vision_feature_layer": [
+    -4,
+    -3,
+    -2,
+    -1
+  ],
+  "vision_feature_select_strategy": "full"
+}

generation_config.json

@@ -0,0 +1,10 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 0,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 0,
+  "transformers_version": "5.8.1",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:736d1d4ff4f960d1c876502fce575b1d950ec1bc6e28dd36fc9ae6eecfbedd80
+size 13835872

processor_config.json

@@ -0,0 +1,50 @@
+{
+  "image_processor": {
+    "crop_size": {
+      "height": 112,
+      "width": 112
+    },
+    "do_center_crop": true,
+    "do_convert_rgb": null,
+    "do_normalize": true,
+    "do_pad": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_grid_pinpoints": [
+      [
+        112,
+        112
+      ],
+      [
+        112,
+        224
+      ],
+      [
+        224,
+        112
+      ]
+    ],
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "LlavaNextImageProcessor",
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "height": 112,
+      "width": 112
+    }
+  },
+  "image_token": "<image>",
+  "num_additional_image_tokens": 0,
+  "patch_size": 14,
+  "processor_class": "LlavaNextProcessor",
+  "vision_feature_select_strategy": "full"
+}

tokenizer_config.json

@@ -0,0 +1,22 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": "<|end_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|end_of_text|>",
+  "errors": "replace",
+  "extra_special_tokens": [
+    "<|start_of_role|>",
+    "<|end_of_role|>",
+    "<|tool_call|>"
+  ],
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 131072,
+  "pad_token": "<|end_of_text|>",
+  "padding_side": "right",
+  "processor_class": "LlavaNextProcessor",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": "<|end_of_text|>",
+  "vocab_size": 49152
+}