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

README.md

@@ -0,0 +1,63 @@
+---
+license: apache-2.0
+base_model: friendshipkim/Qwen2.5-Math-1.5B
+tags:
+- qwen2
+- reward-model
+- success-rate-prediction
+- custom
+---
+
+# Qwen2.5-Math-1.5B-Scoring
+
+This is a custom Qwen2 model with dual heads:
+1. **Language Model Head**: Standard next-token prediction for text generation
+2. **Success Rate Head**: Predicts a success probability score in [0, 1] for the sequence
+
+## Base Model
+
+This model is based on [friendshipkim/Qwen2.5-Math-1.5B](https://huggingface.co/friendshipkim/Qwen2.5-Math-1.5B).
+
+## Usage
+
+```python
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+# Load model with trust_remote_code=True
+model = AutoModelForCausalLM.from_pretrained(
+    "friendshipkim/Qwen2.5-Math-1.5B-Scoring",
+    trust_remote_code=True,
+    torch_dtype="auto"
+)
+tokenizer = AutoTokenizer.from_pretrained("friendshipkim/Qwen2.5-Math-1.5B-Scoring")
+
+# Example: Get both LM output and success score
+prompt = "Question: What is 2+2?\nAnswer: 4"
+inputs = tokenizer(prompt, return_tensors="pt")
+
+# Get both outputs
+lm_output, success_score = model(**inputs, return_score=True)
+print(f"Success rate: {success_score.item():.3f}")
+
+# Generate text (return_score=False for standard generation)
+generated = model.generate(**inputs, max_length=50, return_score=False)
+print(tokenizer.decode(generated[0]))
+```
+
+## Model Architecture
+
+- **Backbone**: Qwen2 transformer model
+- **LM Head**: Linear layer for next-token prediction (vocab_size outputs)
+- **Success Rate Head**: Linear layer for sequence scoring (1 output, sigmoid activation)
+
+## Training
+
+The success_rate_head is randomly initialized and needs to be fine-tuned on your task.
+The LM head and backbone are initialized from the base model.
+
+## Custom Modeling
+
+This model uses a custom modeling file (`modeling_custom.py`) that extends `Qwen2ForCausalLM`.
+The `return_score` parameter controls whether to compute the success rate:
+- `return_score=True`: Returns `(lm_output, success_score)`
+- `return_score=False`: Returns `lm_output` only (for standard generation)

added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

chat_template.jinja

@@ -0,0 +1,54 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- messages[0]['content'] }}
+    {%- else %}
+        {{- 'Please reason step by step, and put your final answer within \\boxed{}.' }}
+    {%- endif %}
+    {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
+    {%- else %}
+        {{- '<|im_start|>system\nPlease reason step by step, and put your final answer within \\boxed{}.<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role }}
+        {%- if message.content %}
+            {{- '\n' + message.content }}
+        {%- endif %}
+        {%- for tool_call in message.tool_calls %}
+            {%- if tool_call.function is defined %}
+                {%- set tool_call = tool_call.function %}
+            {%- endif %}
+            {{- '\n<tool_call>\n{"name": "' }}
+            {{- tool_call.name }}
+            {{- '", "arguments": ' }}
+            {{- tool_call.arguments | tojson }}
+            {{- '}\n</tool_call>' }}
+        {%- endfor %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

config.json

@@ -0,0 +1,62 @@
+{
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoModelForCausalLM": "modeling_custom.Qwen2ForCausalLMWithReward"
+  },
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 32768,
+  "max_window_layers": 21,
+  "model_type": "qwen2",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.54.1",
+  "use_cache": true,
+  "use_mrope": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "max_new_tokens": 2048,
+  "transformers_version": "4.54.1"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7aa8d665ab44d3b461bed742fe93cdcedc15d2e31e50dc372c126e6f720857b
+size 3087467144

modeling_custom.py

@@ -0,0 +1,155 @@
+"""
+Custom Qwen2 model with parallel LM head and success rate prediction head.
+This file is designed to be pushed to HuggingFace Hub and loaded with trust_remote_code=True.
+
+Usage:
+    from transformers import AutoTokenizer, AutoModelForCausalLM
+    
+    model = AutoModelForCausalLM.from_pretrained(
+        "your-username/your-model-name",
+        trust_remote_code=True
+    )
+    tokenizer = AutoTokenizer.from_pretrained("your-username/your-model-name")
+    
+    # Get both LM output and success score
+    inputs = tokenizer("Question: What is 2+2?\nAnswer: 4", return_tensors="pt")
+    lm_output, success_score = model(**inputs, return_score=True)
+"""
+
+from typing import Optional, Union
+import torch
+from torch import nn
+from transformers import Qwen2ForCausalLM, Qwen2PreTrainedModel
+from transformers.cache_utils import Cache
+from transformers.modeling_outputs import CausalLMOutputWithPast, BaseModelOutputWithPast
+from transformers.processing_utils import Unpack
+from transformers.utils import TransformersKwargs
+
+
+class Qwen2ForCausalLMWithReward(Qwen2ForCausalLM):
+    """
+    Qwen2 Model with both LM head and success rate prediction head in parallel.
+    Can generate text AND score sequences simultaneously.
+    
+    The success_rate_head predicts a scalar score in the range [0, 1] based on
+    the last hidden state of the sequence.
+    """
+
+    def __init__(self, config):
+        super().__init__(config)
+        # Add success rate prediction head alongside existing lm_head
+        self.success_rate_head = nn.Linear(config.hidden_size, 1, bias=False)
+        
+        # Initialize the new head
+        self.post_init()
+
+    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[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        return_score: bool = True,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> Union[CausalLMOutputWithPast, tuple]:
+        """
+        Forward pass with both LM head and success rate prediction head.
+        
+        Args:
+            input_ids: Input token IDs
+            attention_mask: Attention mask
+            position_ids: Position IDs
+            past_key_values: Past key values for caching
+            inputs_embeds: Input embeddings (alternative to input_ids)
+            labels: Labels for language modeling loss
+            use_cache: Whether to use KV cache
+            cache_position: Cache position for generation
+            logits_to_keep: Number of logits to keep (for efficiency)
+            return_score: If True, returns (lm_output, score). If False, returns only lm_output.
+            **kwargs: Additional keyword arguments
+        
+        Returns:
+            If return_score=True: tuple of (CausalLMOutputWithPast, success_scores)
+                - CausalLMOutputWithPast: Standard LM output with logits, loss, etc.
+                - success_scores: torch.Tensor of shape (batch_size,) with values in [0, 1]
+            If return_score=False: CausalLMOutputWithPast only
+        
+        Example:
+            >>> from transformers import AutoTokenizer, AutoModelForCausalLM
+            >>> model = AutoModelForCausalLM.from_pretrained("your-model", trust_remote_code=True)
+            >>> tokenizer = AutoTokenizer.from_pretrained("your-model")
+            >>> 
+            >>> prompt = "Question: What is 2+2?\nAnswer: 4"
+            >>> inputs = tokenizer(prompt, return_tensors="pt")
+            >>> 
+            >>> # Get both outputs
+            >>> lm_output, success_score = model(**inputs, return_score=True)
+            >>> print(f"Success rate: {success_score.item():.3f}")
+            >>> 
+            >>> # Or just LM output
+            >>> lm_output = model(**inputs, return_score=False)
+        """
+        # Get model outputs (backbone forward pass)
+        outputs: BaseModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        
+        # Compute LM logits (same as parent class)
+        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, :])
+
+        # Compute loss if labels are provided
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        # Prepare LM output
+        lm_output = CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+        
+        if not return_score:
+            return lm_output
+        
+        # Compute success rate score from the last non-padding token
+        if attention_mask is not None:
+            batch_size = hidden_states.shape[0]
+            # Find the last non-padding token for each sequence
+            sequence_lengths = attention_mask.sum(dim=1) - 1  # -1 for 0-indexing
+            # Gather the last token hidden state for each sequence
+            pooled_hidden_states = hidden_states[
+                torch.arange(batch_size, device=hidden_states.device), 
+                sequence_lengths
+            ]
+        else:
+            # If no mask, use the last token
+            pooled_hidden_states = hidden_states[:, -1, :]
+        
+        # Compute success rate and apply sigmoid to get [0, 1] range
+        score_logits = self.success_rate_head(pooled_hidden_states)  # (batch_size, 1)
+        success_scores = torch.sigmoid(score_logits).squeeze(-1)  # (batch_size,)
+        
+        return lm_output, success_scores
+
+
+# For AutoModel registration
+AutoModelForCausalLM = Qwen2ForCausalLMWithReward
+

special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c5ae00e602b8860cbd784ba82a8aa14e8feecec692e7076590d014d7b7fdafa
+size 11421896

tokenizer_config.json

@@ -0,0 +1,207 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}