Update Readme

README.md

@@ -1,199 +1,204 @@
+
 ---
-library_name: transformers
-tags: []
+library_name: peft
+base_model: GSAI-ML/LLaDA-8B-Instruct
+tags:
+- text-to-sql
+- diffusion
+- llada
+- lora
+- qlora
+- generated_from_trainer
+datasets:
+- gretelai/synthetic_text_to_sql
+license: apache-2.0
+language:
+- en
 ---
 
-# Model Card for Model ID
-
-<!-- Provide a quick summary of what the model is/does. -->
-
-
-
-## Model Details
-
-### Model Description
-
-<!-- Provide a longer summary of what this model is. -->
-
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
-
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-
-### Model Sources [optional]
-
-<!-- Provide the basic links for the model. -->
-
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
-
-## Uses
-
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-
-### Direct Use
-
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-
-[More Information Needed]
-
-### Downstream Use [optional]
+# LLaDA-8B Text-to-SQL (Diffusion-based)
 
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+## Model Summary
 
-[More Information Needed]
+This model is a **Text-to-SQL** adapter fine-tuned on the `GSAI-ML/LLaDA-8B-Instruct` base model. Unlike traditional Autoregressive (AR) models that generate tokens left-to-right, this model uses **Masked Iterative Generation (Diffusion)**.
 
-### Out-of-Scope Use
+It treats text generation as a diffusion process: starting with a fully masked sequence and iteratively refining/unmasking tokens based on confidence scores. This allows for bi-directional context utilization during generation.
 
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+- **Task:** Text-to-SQL (Converting natural language questions + schema into SQL queries).
+- **Method:** LLaDA (Large Language Diffusion with Autoregression) with Block Diffusion Sampling.
+- **Fine-Tuning:** QLoRA (4-bit Quantization + LoRA).
 
-[More Information Needed]
-
-## Bias, Risks, and Limitations
-
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-
-[More Information Needed]
-
-### Recommendations
-
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-
-## How to Get Started with the Model
-
-Use the code below to get started with the model.
+## Model Details
 
-[More Information Needed]
+- **Developed by:** [Tahamajs/Organization]
+- **Base Model:** [GSAI-ML/LLaDA-8B-Instruct](https://huggingface.co/GSAI-ML/LLaDA-8B-Instruct)
+- **Dataset:** [gretelai/synthetic_text_to_sql](https://huggingface.co/datasets/gretelai/synthetic_text_to_sql) (Subset of 20k samples)
+- **Language:** English (Natural Language) -> SQL
+- **Generation Strategy:** Semi-Autoregressive / Block Diffusion
+
+## How to Use (Inference Code)
+
+**Note:** This model does *not* work with the standard `model.generate()` function because it requires a custom diffusion sampling loop. Use the code below to generate SQL queries.
+
+### 1. Setup & Loading
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+from peft import PeftModel, PeftConfig
+
+# Device setup
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# 1. Load Base Model (4-bit)
+base_model_id = "GSAI-ML/LLaDA-8B-Instruct"
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.float16,
+    bnb_4bit_quant_type="nf4",
+)
+
+model = AutoModelForCausalLM.from_pretrained(
+    base_model_id,
+    quantization_config=bnb_config,
+    device_map="auto",
+    trust_remote_code=True,
+    use_cache=False
+)
+tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True)
+
+# 2. Load LoRA Adapter (This Repo)
+adapter_model_id = "YOUR_USERNAME/llada-text-to-sql-lora" # Replace with your repo name
+model = PeftModel.from_pretrained(model, adapter_model_id)
+model.eval()
+
+```
+
+### 2. Define Block Diffusion Generation
+
+```python
+@torch.no_grad()
+def generate_block_diffusion(model, tokenizer, prompt_text, steps=32, gen_len=64):
+    """
+    Generates text using LLaDA's block diffusion strategy.
+    """
+    # Tokenize Prompt
+    prompt_ids = tokenizer.encode(prompt_text, return_tensors='pt').to(model.device)
+    prompt_len = prompt_ids.shape[1]
+    
+    # Initialize Response with [MASK] tokens
+    mask_ids = torch.full((1, gen_len), tokenizer.mask_token_id, device=model.device)
+    input_ids = torch.cat([prompt_ids, mask_ids], dim=1)
+    
+    # Track unknown indices (initially all response tokens)
+    unknown_indices = set(range(prompt_len, input_ids.shape[1]))
+    tokens_to_lock_per_step = gen_len // steps
+
+    for step in range(steps):
+        # Forward pass
+        outputs = model(input_ids)
+        probs = torch.softmax(outputs.logits, dim=-1)
+        
+        # Get most confident predictions
+        confidences, predicted_ids = torch.max(probs, dim=-1)
+        
+        # Identify which tokens to "lock in" this step
+        candidates = []
+        current_unknowns = list(unknown_indices)
+        if not current_unknowns: break
+        
+        for idx in current_unknowns:
+            score = confidences[0, idx].item()
+            token = predicted_ids[0, idx].item()
+            candidates.append((score, idx, token))
+            
+        # Sort by confidence and pick top k
+        candidates.sort(key=lambda x: x[0], reverse=True)
+        top_k = candidates[:tokens_to_lock_per_step]
+        
+        # Update input_ids
+        for _, idx, token in top_k:
+            input_ids[0, idx] = token
+            unknown_indices.remove(idx)
+            
+    # Decode only the generated part
+    return tokenizer.decode(input_ids[0, prompt_len:], skip_special_tokens=True)
+
+```
+
+### 3. Run Inference
+
+```python
+schema = "CREATE TABLE users (id INTEGER, name TEXT, age INTEGER);"
+question = "Show me the names of users older than 25."
+
+prompt = f"""
+<|im_start|>system
+You are a Text-to-SQL assistant. Output ONLY the SQL query. Do not add explanations.<|im_end|>
+<|im_start|>user
+Schema:
+{schema}
+
+Question:
+{question}<|im_end|>
+<|im_start|>assistant
+"""
+
+output = generate_block_diffusion(model, tokenizer, prompt, steps=32, gen_len=64)
+print("Generated SQL:", output)
+
+```
 
 ## Training Details
 
-### Training Data
-
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-
-[More Information Needed]
-
-### Training Procedure
-
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-
-#### Preprocessing [optional]
-
-[More Information Needed]
-
-
-#### Training Hyperparameters
+### Training Configuration
 
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+* **Epochs:** 5
+* **Batch Size:** 2 (Effective Batch Size = 8 via Gradient Accumulation)
+* **Optimizer:** AdamW (lr=2e-4)
+* **Scheduler:** Linear with Warmup (50 steps)
+* **Context Length:** 384 tokens
+* **Precision:** fp16 (via Mixed Precision)
 
-#### Speeds, Sizes, Times [optional]
+### Noise Schedule
 
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+Training used a **Forward Masking** process where tokens in the answer were randomly replaced with `[MASK]` based on a uniform time step . Loss was calculated only on masked tokens and reweighted by .
 
-[More Information Needed]
+### LoRA Configuration
 
-## Evaluation
+* **Rank (r):** 16
+* **Alpha:** 32
+* **Target Modules:** `q_proj`, `v_proj`
+* **Dropout:** 0.05
 
-<!-- This section describes the evaluation protocols and provides the results. -->
+## Evaluation Results
 
-### Testing Data, Factors & Metrics
+Evaluated on the `gretelai/synthetic_text_to_sql` test set (200 samples) using Block Diffusion sampling.
 
-#### Testing Data
+| Metric | Score |
+| --- | --- |
+| **Exact Match (EM)** | ~30% |
+| **Normalized EM** | ~35-40%* |
 
-<!-- This should link to a Dataset Card if possible. -->
+**Scores may vary depending on post-processing strictness and SQL normalization logic.*
 
-[More Information Needed]
-
-#### Factors
-
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-
-[More Information Needed]
-
-#### Metrics
-
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-
-[More Information Needed]
-
-### Results
-
-[More Information Needed]
-
-#### Summary
-
-
-
-## Model Examination [optional]
-
-<!-- Relevant interpretability work for the model goes here -->
-
-[More Information Needed]
-
-## Environmental Impact
-
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-
-## Technical Specifications [optional]
-
-### Model Architecture and Objective
-
-[More Information Needed]
-
-### Compute Infrastructure
-
-[More Information Needed]
-
-#### Hardware
-
-[More Information Needed]
-
-#### Software
-
-[More Information Needed]
-
-## Citation [optional]
-
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-
-**BibTeX:**
-
-[More Information Needed]
-
-**APA:**
-
-[More Information Needed]
-
-## Glossary [optional]
-
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-
-[More Information Needed]
+## Bias, Risks, and Limitations
 
-## More Information [optional]
+* **"Chatty" Output:** The model sometimes fails to produce an EOS token immediately after the semicolon, occasionally repeating the query or adding conversational filler. Post-processing (regex extraction of `SELECT ... ;`) is recommended.
+* **Hallucination:** In complex queries, the model may occasionally hallucinate columns that do not exist in the provided schema if the schema context is too long or complex.
+* **Inference Speed:** Due to the iterative nature of Block Diffusion (multiple forward passes per generation), inference is slower than standard Autoregressive models of the same size.
 
-[More Information Needed]
+## Citation
 
-## Model Card Authors [optional]
+If you use this model or the LLaDA technique, please cite the original paper:
 
-[More Information Needed]
+```bibtex
+@article{nie2024llada,
+  title={LLaDA: Large Language Diffusion with Autoregression},
+  author={Nie, Shen and others},
+  journal={arXiv preprint arXiv:2402.XXXXX},
+  year={2024}
+}
 
-## Model Card Contact
+```
 
-[More Information Needed]
+```