uploaded weights

weights_for_huggingface/README.md

@@ -0,0 +1,109 @@
+---
+language:
+- en
+tags:
+- regression
+- similarity
+- sql
+- natural-language
+- reward-model
+license: mit
+datasets:
+- custom
+metrics:
+- mse
+- mae
+- rmse
+model-index:
+- name: BERT Reward Model for CoT Filtering
+  results:
+  - task:
+      type: regression
+      name: Similarity Score Prediction
+    dataset:
+      name: Custom CoT Dataset
+      type: custom
+    metrics:
+    - type: mse
+      value: 0.0238
+    - type: mae
+      value: 0.1229
+    - type: rmse
+      value: 0.1543
+---
+
+# BERT Reward Model for CoT Filtering
+
+A BERT-based regression model fine-tuned to predict similarity scores between SQL queries, reasoning chains (Chain-of-Thought), and natural language descriptions.
+
+## Model Description
+
+This model is based on `bert-base-uncased` and has been fine-tuned for regression to predict similarity scores in the range [0, 1]. The model takes as input a concatenation of:
+- SQL query
+- Reasoning/Chain-of-Thought explanation
+- Predicted natural language description
+
+And outputs a similarity score indicating how well the predicted NL matches the ground truth.
+
+## Usage
+
+```python
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+import torch
+
+# Load model and tokenizer
+model_path = "path/to/weights_for_huggingface"
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+model = AutoModelForSequenceClassification.from_pretrained(
+    model_path,
+    num_labels=1,
+    problem_type="regression"
+)
+model.eval()
+
+# Prepare input
+sql = "SELECT movie_title FROM movies WHERE movie_release_year = 1945"
+reasoning = "think: The SQL selects the movie title..."
+predicted_nl = "What was the most popular movie released in 1945?"
+
+input_text = f"SQL: {sql}\nReasoning: {reasoning}\nNL: {predicted_nl}"
+
+# Tokenize and predict
+inputs = tokenizer(input_text, return_tensors="pt", truncation=True, max_length=512)
+with torch.no_grad():
+    outputs = model(**inputs)
+    # Apply sigmoid to get probability
+    similarity_score = torch.sigmoid(outputs.logits).item()
+
+print(f"Predicted similarity: {similarity_score:.3f}")
+```
+
+## Training Details
+
+- **Base Model**: bert-base-uncased
+- **Training Dataset**: Custom CoT dataset with corruptions (7,342 examples)
+- **Train/Val/Test Split**: 75% / 12.5% / 12.5%
+- **Training Loss**: MSE (Mean Squared Error)
+- **Evaluation Metrics**:
+  - MSE: 0.0238
+  - MAE: 0.1229
+  - RMSE: 0.1543
+
+## Limitations
+
+- Maximum input length: 512 tokens (BERT's limit)
+- Trained on a specific domain (SQL to NL translation with CoT)
+- Performance may vary on out-of-domain data
+
+## Citation
+
+If you use this model, please cite:
+
+```bibtex
+@misc{bert_cot_reward_model,
+  title={BERT Reward Model for Chain-of-Thought Filtering},
+  author={Your Name},
+  year={2024},
+}
+```
+

weights_for_huggingface/config.json

@@ -0,0 +1,28 @@
+{
+  "architectures": [
+    "BertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "id2label": null,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "label2id": null,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_labels": 1,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "problem_type": "regression",
+  "transformers_version": "4.57.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

weights_for_huggingface/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:844ef2517807f6ac2ea74068e7ee6d178ae89eaf02c08ef2aa531cf2f020d433
+size 437955572

weights_for_huggingface/special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

weights_for_huggingface/tokenizer_config.json

@@ -0,0 +1,56 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "[CLS]",
+  "do_lower_case": true,
+  "extra_special_tokens": {},
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}