Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

Modelfile

@@ -0,0 +1,14 @@
+# ollama modelfile auto-generated by llamafactory
+
+FROM .
+
+TEMPLATE """<bos>{{ if .System }}{{ .System }}
+
+{{ end }}{{ range .Messages }}{{ if eq .Role "user" }}<start_of_turn>user
+{{ .Content }}<end_of_turn>
+<start_of_turn>model
+{{ else if eq .Role "assistant" }}{{ .Content }}<end_of_turn>
+{{ end }}{{ end }}"""
+
+PARAMETER stop "<end_of_turn>"
+PARAMETER num_ctx 4096

README.md

@@ -1,3 +1,85 @@
----
-license: apache-2.0
----
+# Gemma-3n-E2B-it Android Control LoRA Fine-tuned Model
+
+## Model Overview
+This model is a fine-tuned version of Google's `gemma-3n-E2B-it` base model with LoRA adaptation for Android UI control tasks.
+
+## Training Data
+- **Dataset**: [OfficerChul/gemma-3n-E2B-it-Antroid-Control-84k](https://huggingface.co/OfficerChul/gemma-3n-E2B-it-Antroid-Control-84k)
+- **Data Format**: Mobile UI screenshots paired with user instructions to perform appropriate actions (click, scroll, input, etc.)
+
+### Training Data Format Example
+```json
+{
+  "messages": [
+    {
+      "role": "system",
+      "content": "You are a helpful assistant that can identify what action to perform on mobile UI Screenshot given the user instruction."
+    },
+    {
+      "role": "user",
+      "content": "<image>Click on the Recording 2"
+    },
+    {
+      "role": "assistant",
+      "content": "{\"action_type\": \"click\", \"x\": 561, \"y\": 535}"
+    }
+  ],
+  "images": ["and_ctrl/out_episode_18557_step_001.png"]
+}
+```
+
+## Training Method
+LoRA fine-tuning performed using [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory) framework
+
+### 1. Training Configuration (`gemma3n-e2b-it.yaml`)
+- **Base Model**: `google/gemma-3n-E2B-it`
+- **Training Method**: LoRA (Low-Rank Adaptation)
+- **LoRA Configuration**:
+  - Rank: 32
+  - Target modules: `q_proj, k_proj, v_proj, o_proj`
+- **Training Parameters**:
+  - Batch size: 4 (gradient accumulation: 48)
+  - Learning rate: 2e-5
+  - Epochs: 5
+  - LR scheduler: Cosine
+  - Optimizer: AdamW (fused)
+  - Precision: bf16
+- **Additional Settings**:
+  - Gradient checkpointing enabled
+  - Vision tower, multi-modal projector, and language model all trainable
+  - DeepSpeed ZeRO-2 utilized
+
+### 2. Model Merging (`gemma3n-e2b-it_lora_sft_merge.yaml`)
+Merged trained LoRA adapter with base model:
+- **Base Model**: `google/gemma-3n-E2B-it`
+
+## Supported Action Types
+- `click`: Click on specific coordinates
+- `long_press`: Long press action
+- `scroll`: Scroll (up/down)
+- `input_text`: Text input
+- `navigate_back`: Navigate back
+- `navigate_home`: Navigate to home screen
+- `open_app`: Open application
+- `wait`: Wait action
+
+## Usage
+The merged model can be directly loaded using the Hugging Face Transformers library.
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model_path = "output/gemma-e2b-lora-sft"
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+model = AutoModelForCausalLM.from_pretrained(
+    model_path,
+    trust_remote_code=True
+)
+```
+
+## License
+Follows the license terms of the Google Gemma model.
+
+## Notes
+- This model was developed for research purposes in mobile UI automation and accessibility enhancement
+- Proper validation is required when using in production environments

chat_template.jinja

@@ -0,0 +1,49 @@
+{{ bos_token }}
+{%- if messages[0]['role'] == 'system' -%}
+    {%- if messages[0]['content'] is string -%}
+        {%- set first_user_prefix = messages[0]['content'] + '
+
+' -%}
+    {%- else -%}
+        {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
+
+' -%}
+    {%- endif -%}
+    {%- set loop_messages = messages[1:] -%}
+{%- else -%}
+    {%- set first_user_prefix = "" -%}
+    {%- set loop_messages = messages -%}
+{%- endif -%}
+{%- for message in loop_messages -%}
+    {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
+        {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
+    {%- endif -%}
+    {%- if (message['role'] == 'assistant') -%}
+        {%- set role = "model" -%}
+    {%- else -%}
+        {%- set role = message['role'] -%}
+    {%- endif -%}
+    {{ '<start_of_turn>' + role + '
+' + (first_user_prefix if loop.first else "") }}
+    {%- if message['content'] is string -%}
+        {{ message['content'] | trim }}
+    {%- elif message['content'] is iterable -%}
+        {%- for item in message['content'] -%}
+            {%- if item['type'] == 'audio' -%}
+                {{ '<audio_soft_token>' }}
+            {%- elif item['type'] == 'image' -%}
+                {{ '<image_soft_token>' }}
+            {%- elif item['type'] == 'text' -%}
+                {{ item['text'] | trim }}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{ raise_exception("Invalid content type") }}
+    {%- endif -%}
+    {{ '<end_of_turn>
+' }}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    {{'<start_of_turn>model
+'}}
+{%- endif -%}

config.json

@@ -0,0 +1,210 @@
+{
+  "architectures": [
+    "Gemma3nForConditionalGeneration"
+  ],
+  "audio_config": {
+    "conf_attention_chunk_size": 12,
+    "conf_attention_context_left": 13,
+    "conf_attention_context_right": 0,
+    "conf_attention_logit_cap": 50.0,
+    "conf_conv_kernel_size": 5,
+    "conf_num_attention_heads": 8,
+    "conf_num_hidden_layers": 12,
+    "conf_reduction_factor": 4,
+    "conf_residual_weight": 0.5,
+    "dtype": "bfloat16",
+    "gradient_clipping": 10000000000.0,
+    "hidden_size": 1536,
+    "input_feat_size": 128,
+    "model_type": "gemma3n_audio",
+    "rms_norm_eps": 1e-06,
+    "sscp_conv_channel_size": [
+      128,
+      32
+    ],
+    "sscp_conv_group_norm_eps": 0.001,
+    "sscp_conv_kernel_size": [
+      [
+        3,
+        3
+      ],
+      [
+        3,
+        3
+      ]
+    ],
+    "sscp_conv_stride_size": [
+      [
+        2,
+        2
+      ],
+      [
+        2,
+        2
+      ]
+    ],
+    "vocab_offset": 262272,
+    "vocab_size": 128
+  },
+  "audio_soft_tokens_per_image": 188,
+  "audio_token_id": 262273,
+  "boa_token_id": 256000,
+  "boi_token_id": 255999,
+  "dtype": "bfloat16",
+  "eoa_token_id": 262272,
+  "eoi_token_id": 262144,
+  "eos_token_id": [
+    1,
+    106
+  ],
+  "hidden_size": 2048,
+  "image_token_id": 262145,
+  "initializer_range": 0.02,
+  "model_type": "gemma3n",
+  "text_config": {
+    "activation_sparsity_pattern": [
+      0.95,
+      0.95,
+      0.95,
+      0.95,
+      0.95,
+      0.95,
+      0.95,
+      0.95,
+      0.95,
+      0.95,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0,
+      0.0
+    ],
+    "altup_active_idx": 0,
+    "altup_coef_clip": 120.0,
+    "altup_correct_scale": true,
+    "altup_num_inputs": 4,
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "dtype": "bfloat16",
+    "final_logit_softcapping": 30.0,
+    "head_dim": 256,
+    "hidden_activation": "gelu_pytorch_tanh",
+    "hidden_size": 2048,
+    "hidden_size_per_layer_input": 256,
+    "initializer_range": 0.02,
+    "intermediate_size": [
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192,
+      8192
+    ],
+    "laurel_rank": 64,
+    "layer_types": [
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention"
+    ],
+    "max_position_embeddings": 32768,
+    "model_type": "gemma3n_text",
+    "num_attention_heads": 8,
+    "num_hidden_layers": 30,
+    "num_key_value_heads": 2,
+    "num_kv_shared_layers": 10,
+    "rms_norm_eps": 1e-06,
+    "rope_local_base_freq": 10000.0,
+    "rope_scaling": null,
+    "rope_theta": 1000000.0,
+    "sliding_window": 512,
+    "use_cache": true,
+    "vocab_size": 262400,
+    "vocab_size_per_layer_input": 262144
+  },
+  "transformers_version": "4.57.0",
+  "use_cache": true,
+  "vision_config": {
+    "architecture": "mobilenetv5_300m_enc",
+    "do_pooling": false,
+    "dtype": "bfloat16",
+    "hidden_size": 2048,
+    "initializer_range": 0.02,
+    "label_names": [
+      "LABEL_0",
+      "LABEL_1"
+    ],
+    "model_args": null,
+    "model_type": "gemma3n_vision",
+    "num_classes": 2,
+    "rms_norm_eps": 1e-06,
+    "vocab_offset": 262144,
+    "vocab_size": 128
+  },
+  "vision_soft_tokens_per_image": 256
+}

gemma3n-e2b-it.yaml

@@ -0,0 +1,57 @@
+### model
+model_name_or_path: google/gemma-3n-E2B-it
+image_max_pixels: 1048576
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: lora
+freeze_vision_tower: false
+freeze_multi_modal_projector: false
+freeze_language_model: false
+lora_rank: 32
+lora_target: "q_proj, k_proj, v_proj, o_proj"
+deepspeed: examples/deepspeed/ds_z2_config.json
+
+### dataset
+dataset:  and_ctrl_train # video: mllm_video_demo
+template: gemma3n
+cutoff_len: 1024               # 2048 -> 1024 (토큰 짧으면 속도/메모리 이득)
+# max_samples: 1000            # 전체 학습이면 제거/주석 처리
+overwrite_cache: true
+preprocessing_num_workers: 8
+dataloader_num_workers: 8     # 8~16 권장, 중간값으로 상향
+
+### output
+output_dir: saves/gemma-e2b/lora/sft/train
+logging_steps: 1
+save_steps: 300
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: false
+report_to: wandb  # choices: [none, wandb, tensorboard, swanlab, mlflow]
+run_name: and_ctrl-skt-gemma-e2b-lora-sft-train
+
+### train
+per_device_train_batch_size: 4
+gradient_accumulation_steps: 48
+learning_rate: 2e-5
+num_train_epochs: 5.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+resume_from_checkpoint: null
+seed: 42
+
+# 추가 적용(요청 사항)
+gradient_checkpointing: true
+optim: adamw_torch_fused
+weight_decay: 0.01
+
+### eval
+val_size: 0.05
+per_device_eval_batch_size: 1
+eval_strategy: steps
+eval_steps: 500

gemma3n-e2b-it_lora_sft_merge.yaml

@@ -0,0 +1,11 @@
+### model
+model_name_or_path: google/gemma-3n-E2B-it
+adapter_name_or_path: saves/gemma-e2b/lora/sft/train
+template: gemma3n
+trust_remote_code: true
+
+### export
+export_dir: output/gemma-e2b-lora-sft
+export_size: 5
+export_device: auto
+export_legacy_format: false

generation_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "do_sample": true,
+  "eos_token_id": [
+    1,
+    106
+  ],
+  "pad_token_id": 0,
+  "top_k": 64,
+  "top_p": 0.95,
+  "transformers_version": "4.57.0"
+}

model-00001-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a1a78f91baca05bb933d67e3dad9e643b56b5b45c3384463879c0ed59a3a8e4
+size 4983256208

model-00002-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af670d3e63e9afd9feb88a3f044bc9b6a046a18b232b5a054f1c912c68d21fe8
+size 4992754064

model-00003-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f6c2f72292cb7b0b2081dde3451cbe423e47288dd8409f851f3785c3ce38bf0
+size 903075744

preprocessor_config.json

@@ -0,0 +1,53 @@
+{
+  "crop_size": null,
+  "data_format": "channels_first",
+  "default_to_square": false,
+  "device": null,
+  "disable_grouping": null,
+  "dither": 0.0,
+  "do_center_crop": null,
+  "do_convert_rgb": null,
+  "do_normalize": false,
+  "do_pad": null,
+  "do_rescale": true,
+  "do_resize": true,
+  "feature_size": 128,
+  "fft_length": 1024,
+  "fft_overdrive": true,
+  "frame_length": 512,
+  "hop_length": 160,
+  "image_mean": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "image_processor_type": "SiglipImageProcessorFast",
+  "image_seq_length": 256,
+  "image_std": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "input_data_format": null,
+  "input_scale_factor": 1.0,
+  "max_frequency": 7600.0,
+  "mel_floor": 1e-05,
+  "min_frequency": 125.0,
+  "pad_size": null,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "per_bin_mean": null,
+  "per_bin_stddev": null,
+  "preemphasis": 0.97,
+  "preemphasis_htk_flavor": true,
+  "processor_class": "Gemma3nProcessor",
+  "resample": 2,
+  "rescale_factor": 0.00392156862745098,
+  "return_attention_mask": true,
+  "return_tensors": null,
+  "sampling_rate": 16000,
+  "size": {
+    "height": 768,
+    "width": 768
+  }
+}

processor_config.json

@@ -0,0 +1,5 @@
+{
+  "audio_seq_length": 188,
+  "image_seq_length": 256,
+  "processor_class": "Gemma3nProcessor"
+}

special_tokens_map.json

@@ -0,0 +1,36 @@
+{
+  "audio_token": "<audio_soft_token>",
+  "boa_token": "<start_of_audio>",
+  "boi_token": "<start_of_image>",
+  "bos_token": {
+    "content": "<bos>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eoa_token": "<end_of_audio>",
+  "eoi_token": "<end_of_image>",
+  "eos_token": {
+    "content": "<end_of_turn>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "image_token": "<image_soft_token>",
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6c35ee648c07754b44cd9e371c75d4caa05c4504910b7ad29b1847ee9d8ba5d
+size 33442553

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea5f0cc48abfbfc04d14562270a32e02149a3e7035f368cc5a462786f4a59961
+size 4696020