Update README.md

README.md

@@ -1,15 +1,33 @@
+---
+base_model: gemma-3-27b-it
+tags:
+- text-generation-inference
+- transformers
+- gemma3
+- reasoning
+- mathematics
+- grpo
+license: apache-2.0
+language:
+- en
+inference:
+  parameters:
+    temperature: 0.7
+    top_p: 0.95
+    top_k: 64
+    max_new_tokens: 512
+---
+
 # Gemma 3 27B GRPO Reasoning Model
 
 ## Model Description
 - **Developed by:** klei1
 - **Model type:** Gemma 3 27B fine-tuned with GRPO for reasoning tasks
 - **License:** apache-2.0
-- **Finetuned from model:** [unsloth/gemma-3-27b-it-unsloth-bnb-4bit](https://huggingface.co/unsloth/gemma-3-27b-it-unsloth-bnb-4bit)
-- **Framework:** Hugging Face Transformers with Unsloth optimization
-
-This model is a fine-tuned version of Google's Gemma 3 27B instruction-tuned model, enhanced using Generative Rejection Policy Optimization (GRPO) to improve its reasoning capabilities. The training was completed 1.6x faster with [Unsloth](https://github.com/unslothai/unsloth) optimization, requiring 60% less VRAM and enabling 6x longer context than environments with Flash Attention 2.
+- **Finetuned from model:** Google's Gemma 3 27B instruction-tuned model
+- **Framework:** Hugging Face Transformers
 
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
+This model is a fine-tuned version of Google's Gemma 3 27B instruction-tuned model, enhanced using Generative Rejection Policy Optimization (GRPO) to improve its reasoning capabilities.
 
 ## Capabilities & Training
 
@@ -29,14 +47,17 @@ The model has been trained to follow a specific reasoning format:
 - Final solutions are provided between `<SOLUTION>` and `</SOLUTION>` tags
 
 ### Training Configuration
-- **Framework:** Unsloth with Hugging Face's TRL library
+- **Framework:** Hugging Face's TRL library
 - **Optimization:** LoRA fine-tuning (r=8, alpha=8)
-- **Precision:** 4-bit dynamic quantization for superior accuracy with minimal VRAM usage
 - **Reward Functions:** Format adherence, answer accuracy, and reasoning quality
-- **Context Length:** Up to 128K tokens supported by base model
 
 ## Technical Specifications
 
+### Available Formats
+This model is available in two formats:
+- Standard adapter format (adapter_model.safetensors)
+- GGUF 8-bit quantized format (bleta-meditor-27b-finetune.Q8_0.gguf) for use with llama.cpp
+
 ### Gemma 3 Architecture Benefits
 - 27B parameters, trained on 14 trillion tokens
 - 128K context window (extended from 32K)
@@ -44,15 +65,10 @@ The model has been trained to follow a specific reasoning format:
 - 5 sliding + 1 global attention pattern
 - 1024 sliding window attention
 
-### Unsloth Optimization
-- 1.6x faster training compared to standard implementations
-- >60% VRAM reduction, enabling training on consumer GPUs
-- Support for 6x longer context than environments with Flash Attention 2
-- Fixes for float16 mixed precision issues that cause infinity in activations and gradients
+## System Prompt
 
-## Usage
+To get the best results from this model, use this system prompt:
 
-### Example System Prompt
 ```
 You are given a problem.
 Think about the problem and provide your working out.
@@ -60,43 +76,6 @@ Place it between <start_working_out> and <end_working_out>.
 Then, provide your solution between <SOLUTION></SOLUTION>
 ```
 
-### Example Usage
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-import torch
-
-model_name = "klei1/gemma-3-27b-grpo"  # Replace with your model path
-model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype=torch.float16)
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-
-system_prompt = """You are given a problem.
-Think about the problem and provide your working out.
-Place it between <start_working_out> and <end_working_out>.
-Then, provide your solution between <SOLUTION></SOLUTION>"""
-
-messages = [
-    {"role": "system", "content": system_prompt},
-    {"role": "user", "content": "If a train travels at 60 miles per hour, how far will it travel in 2.5 hours?"}
-]
-
-text = tokenizer.apply_chat_template(
-    messages,
-    add_generation_prompt=True,
-    tokenize=False
-)
-
-inputs = tokenizer(text, return_tensors="pt").to(model.device)
-outputs = model.generate(
-    **inputs,
-    max_new_tokens=512,
-    temperature=0.7,
-    top_p=0.95,
-    top_k=64
-)
-
-print(tokenizer.decode(outputs[0], skip_special_tokens=False))
-```
-
 ## Limitations
 - While this model excels at reasoning tasks, particularly mathematical problems, it may still occasionally provide incorrect solutions for complex problems.
 - The model's performance might vary depending on problem complexity and wording.
@@ -104,7 +83,6 @@ print(tokenizer.decode(outputs[0], skip_special_tokens=False))
 
 ## Acknowledgments
 - Google for developing the Gemma 3 model family
-- Unsloth team for providing optimization techniques that make fine-tuning large models more accessible
 - Hugging Face for their TRL library and GRPO implementation
 
 ## Citation