Update README.md

README.md

@@ -26,65 +26,35 @@ library_name: transformers
 Original model technical report
 
 [Phi-4 Technical Report](https://arxiv.org/pdf/2412.08905)
+# Phi-4-MedIT-10B-o1 🚀
 
-## Model Summary 
+Phi-4-MedIT-10B-o1 is a specialized large language model pruned for efficiency and performance:
 
-|                         |                                                                               |     
-|-------------------------|-------------------------------------------------------------------------------|
-| **Developers**          | Microsoft Research                                                            |
-| **Description**         | `phi-4` is a state-of-the-art open model built upon a blend of synthetic datasets, data from filtered public domain websites, and acquired academic books and Q&A datasets. The goal of this approach was to ensure that small capable models were trained with data focused on high quality and advanced reasoning.<br><br>`phi-4` underwent a rigorous enhancement and alignment process, incorporating both supervised fine-tuning and direct preference optimization to ensure precise instruction adherence and robust safety measures                |
-| **Architecture**        | 14B parameters, dense decoder-only Transformer model                          |
-| **Inputs**              | Text, best suited for prompts in the chat format                              |
-| **Context length**      | 16K tokens                                                                    |
-| **GPUs**                | 1920 H100-80G                                                                 |
-| **Training time**       | 21 days                                                                       |
-| **Training data**       | 9.8T tokens                                                                   |
-| **Outputs**             | Generated text in response to input                                           |
-| **Dates**               | October 2024 – November 2024                                                  |
-| **Status**              | Static model trained on an offline dataset with cutoff dates of June 2024 and earlier for publicly available data                                                                               |
-| **Release date**        | December 12, 2024                                                             |
-| **License**             | MIT                                                                         |
+### Key Features:
 
-## Intended Use 
+- **Pruned Model:**
+  - Derived from the **phi-4 14B model**, reduced to 10B parameters using the **MKA2G pruning technique**.
+  - For details on the pruning methodology, see [MedITSolutionsKurman/llama-pruning](https://github.com/mkurman/llama-pruning) on GitHub.
 
-|                               |                                                                         |
-|-------------------------------|-------------------------------------------------------------------------|
-| **Primary Use Cases**         | Our model is designed to accelerate research on language models, for use as a building block for generative AI powered features. It provides uses for general purpose AI systems and applications (primarily in English) which require:<br><br>1. Memory/compute constrained environments.<br>2. Latency bound scenarios.<br>3. Reasoning and logic.                                                                       |
-| **Out-of-Scope Use Cases**    | Our models is not specifically designed or evaluated for all downstream purposes, thus:<br><br>1. Developers should consider common limitations of language models as they select use cases, and evaluate and mitigate for accuracy, safety, and fairness before using within a specific downstream use case, particularly for high-risk scenarios.<br>2. Developers should be aware of and adhere to applicable laws or regulations (including privacy, trade compliance laws, etc.) that are relevant to their use case, including the model’s focus on English.<br>3. Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.                                                              |
+- **Reasoning SFT:**
+  - Fine-tuned with a **single epoch of reasoning-specific supervised fine-tuning (SFT)** for optimized reasoning tasks.
 
-## Data Overview 
+### Recommended Usage:
 
-### Training Datasets 
+- **ReasonFlow Integration:**
+  - Achieve advanced reasoning by leveraging the [ReasonFlow framework](https://github.com/mkurman/ReasonFlow), which uses Multi-Path Generation with Noise.
 
-Our training data is an extension of the data used for Phi-3 and includes a wide variety of sources from:
+- **Monte-Carlo Tree Search (MCTS):**
+  - Use with [MCTS-PyTorch](https://github.com/mkurman/mcts-pytorch) to explore tree-based decision-making and inference.
 
-1. Publicly available documents filtered rigorously for quality, selected high-quality educational data, and code.
+### Applications:
 
-2. Newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (science, daily activities, theory of mind, etc.).
+This model is ideal for:
+- Efficient, high-quality reasoning tasks.
+- Scenarios requiring enhanced performance in pruned environments.
+- Integration with advanced frameworks for reasoning or decision-making.
 
-3. Acquired academic books and Q&A datasets.
-
-4. High quality chat format supervised data covering various topics to reflect human preferences on different aspects such as instruct-following, truthfulness, honesty and helpfulness.
-
-Multilingual data constitutes about 8% of our overall data. We are focusing on the quality of data that could potentially improve the reasoning ability for the model, and we filter the publicly available documents to contain the correct level of knowledge.
-
-#### Benchmark datasets 
-
-We evaluated `phi-4` using [OpenAI’s SimpleEval](https://github.com/openai/simple-evals) and our own internal benchmarks to understand the model’s capabilities, more specifically: 
-
-* **MMLU:** Popular aggregated dataset for multitask language understanding.
-
-* **MATH:** Challenging competition math problems.
-
-* **GPQA:** Complex, graduate-level science questions.
-
-* **DROP:** Complex comprehension and reasoning.
-
-* **MGSM:** Multi-lingual grade-school math.
-
-* **HumanEval:** Functional code generation.
-
-* **SimpleQA:** Factual responses.
+Explore the possibilities with Phi-4-MedIT-10B-o1 and push the boundaries of efficient AI reasoning!
 
 ## Safety 
 
@@ -150,6 +120,60 @@ outputs = pipeline(messages, max_new_tokens=128)
 print(outputs[0]["generated_text"][-1])
 ```
 
+### With `ReasonFlow`
+Check it here: [ReasonFlow framework](https://github.com/mkurman/ReasonFlow)
+
+```python
+...
+config = ReasonFlowConfig(
+    num_of_thinkers=NUM_OF_THINKERS,
+    num_of_thoughts=NUM_OF_THOUGHTS,
+    topk_thinkers=TOPK_THINKERS,
+)
+
+index = 1
+
+reason_flow = ReasonFlow(config, model, tokenizer)
+output = reason_flow.generate_thoughts(chat_like_texts[index], 
+                                       max_new_tokens=4096, 
+                                       stream=True,
+                                       device='cuda', 
+                                       torch_dtype='bfloat16',
+                                       do_sample=False, 
+                                       temperature=None, 
+                                       top_p=None, 
+                                       top_k=None, 
+                                       repetition_penalty=None)
+```
+
+### With `Monte-Carlo Tree Search`
+Check it here: [MCTS-PyTorch](https://github.com/mkurman/mcts-pytorch)
+
+```python
+...
+model.to("cuda")
+input_tokens = input_ids['input_ids'].to('cuda') 
+
+model.eval()
+
+print(tokenizer.decode(input_tokens[0]))
+
+while len(input_tokens[0]) < MAX_TOTAL_TOKENS:
+    mcts = MCTS(model, tokenizer, max_depth=MAX_DEPTH, num_simulations=MAX_SIMULATIONS, temperature=TEMPERATURE, max_new_tokens=MAX_NEW_TOKENS, stop_tokens=model.config.eos_token_id)
+    
+    new_tokens = mcts.search(input_tokens)
+
+    new_tokens = new_tokens[..., input_tokens.shape[-1]:]
+
+    input_tokens = torch.cat([input_tokens, new_tokens], dim=-1)
+
+    print(tokenizer.decode(new_tokens[0]), end='')
+
+    if has_eos(new_tokens, eos_token=model.config.eos_token_id):
+        break
+
+```
+
 ## Responsible AI Considerations
 
 Like other language models, `phi-4` can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include: