Update README.md

README.md

@@ -2,285 +2,304 @@
 license: apache-2.0
 language:
 - en
+base_model:
+- Qwen/Qwen3-8B
 library_name: transformers
 pipeline_tag: text-generation
 tags:
 - medical
-- healthcare
-- reasoning
-- llm
-- deepseek
-- qwen
+- clinical-reasoning
+- chain-of-thought
+- qwen3
+- sft
+- dpo
 - tanit
-base_model: deepseek-ai/DeepSeek-R1-0528-Qwen3-8B
+datasets:
+- FreedomIntelligence/medical-o1-reasoning-SFT
+- FreedomIntelligence/Medical-R1-Distill-Data
+- UCSC-VLAA/MedReason
+- UCSC-VLAA/m23k-tokenized
+- Intelligent-Internet/II-Medical-RL
+- hongzhouyu/FineMed-DPO
+- super-dainiu/medagents-benchmark
 model-index:
-- name: Tanit-MedReason-8B
+- name: Tanit-Med-8B
   results:
-  - task:
-      type: question-answering
-      name: Medical QA
-    metrics:
-    - type: accuracy
-      value: 60.3
-      name: Test Accuracy
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: MedQA', type: super-dainiu/medagents-benchmark, config: MedQA, split: test}
+    metrics: [{type: accuracy, value: 65.0, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: PubMedQA', type: super-dainiu/medagents-benchmark, config: PubMedQA, split: test}
+    metrics: [{type: accuracy, value: 68.0, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: MedMCQA', type: super-dainiu/medagents-benchmark, config: MedMCQA, split: test}
+    metrics: [{type: accuracy, value: 57.8, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: MedBullets', type: super-dainiu/medagents-benchmark, config: MedBullets, split: test}
+    metrics: [{type: accuracy, value: 42.9, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: MMLU (medical)', type: super-dainiu/medagents-benchmark, config: MMLU, split: test}
+    metrics: [{type: accuracy, value: 77.5, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: MMLU-Pro (medical)', type: super-dainiu/medagents-benchmark, config: MMLU-Pro, split: test}
+    metrics: [{type: accuracy, value: 47.3, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: MedExQA', type: super-dainiu/medagents-benchmark, config: MedExQA, split: test}
+    metrics: [{type: accuracy, value: 73.3, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: MedXpertQA-R', type: super-dainiu/medagents-benchmark, config: MedXpertQA-R, split: test}
+    metrics: [{type: accuracy, value: 12.1, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: MedXpertQA-U', type: super-dainiu/medagents-benchmark, config: MedXpertQA-U, split: test}
+    metrics: [{type: accuracy, value: 14.6, name: Accuracy}]
+  - task: {type: question-answering, name: Medical QA}
+    dataset: {name: 'MedAgentsBench: AfriMedQA', type: super-dainiu/medagents-benchmark, config: AfrimedQA, split: test}
+    metrics: [{type: accuracy, value: 48.9, name: Accuracy}]
 ---
 
-<div align="center">
+# Tanit-Med-8B
 
-# 🏥 Tanit-MedReason-8B
+*An 8B medical reasoning model that thinks before it answers — and a model card that tells you where it doesn't.*
 
-### Advanced Medical Reasoning Model by TANIT Healthcare Technologies
+Tanit-Med-8B is a full fine-tune of [Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) for clinical multiple-choice reasoning and medical question answering. It was trained in four stages — broad medical SFT, reasoning SFT, DPO, and a short chain-of-thought polish — and evaluated across the ten benchmarks in [MedAgentsBench](https://huggingface.co/datasets/super-dainiu/medagents-benchmark), on both the standard and the hard splits.
 
-[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
-[![Model](https://img.shields.io/badge/Model-8B%20Parameters-green.svg)]()
-[![Phase](https://img.shields.io/badge/Phase-Final%20(v1.0)-orange.svg)]()
+It is named for Tanit, the Carthaginian goddess who watched over the western Mediterranean. The name is a commitment as much as a nod: this model is built by a North African team, and AfriMedQA is a first-class benchmark here, not a footnote.
 
-</div>
+**The short version:** on the standard splits, Tanit-Med-8B is, to our knowledge, the strongest open 8B medical model on MedAgentsBench — a 50.7% macro average, against 34.6% for the next-best 8B peer. On the *hard* splits it is not meaningfully better than any other 8B model, and neither is anyone else. We think both halves of that sentence matter, so both are in this card.
 
 ---
 
-## 🌟 Model Overview
+## At a glance
 
-**Tanit-MedReason-8B** is a specialized medical reasoning model developed by [TANIT Healthcare Technologies](https://tanitai.com).
+| | |
+|---|---|
+| **Base model** | Qwen/Qwen3-8B (dense, 8.2B params) |
+| **Precision** | BF16 |
+| **Context** | 32K native (trained at 8,192-token packed sequences) |
+| **Reasoning** | `<think>` blocks, always on by default (≈99% of responses) |
+| **Language** | English (only language evaluated) |
+| **License** | Apache-2.0 |
+| **Best at** | Multiple-choice clinical QA, USMLE-style vignettes, medical exam reasoning |
+| **Not for** | Diagnosis, treatment, dosing, triage, or anything touching a real patient |
 
-Production-ready medical reasoning model trained through a 4-phase curriculum: Broad SFT → Reasoning SFT → DPO Alignment → MedReason Chain-of-Thought fine-tuning. Achieves strong performance across medical benchmarks with transparent reasoning via <think> tags.
+---
 
-### Key Features
+## Results
 
-- 🧠 **Advanced Medical Reasoning**: Deep chain-of-thought reasoning for complex medical scenarios
-- 🔬 **Evidence-Based Responses**: Trained on validated medical knowledge sources
-- 💭 **Transparent Thinking**: Exposes reasoning process via `<think>` tags
-- ⚡ **Efficient**: 8B parameters optimized for deployment
-- 🛡️ **Safety-Aligned**: DPO-trained for safer, more helpful responses
+All numbers below are **zero-shot, greedy decoding**, using the MedAgentsBench prompt and answer-index extraction. `FULL` is the standard test split; `HARD` is the MedAgentsBench hard subset. Baseline numbers for the other 8B models were produced by us under the same harness.
 
+### Against open 8B medical models
 
-## 🔬 Training Pipeline
+| Benchmark | **Tanit-Med-8B** | DeepSeek-R1-0528-Qwen3-8B | Falcon-H1R-7B | HuatuoGPT-o1-8B | MedReason-8B | Ministral-3-8B-Reasoning |
+|---|---|---|---|---|---|---|
+| MedQA | **65.0** | 44.0 | 28.9 | 29.5 | 28.0 | 27.8 |
+| PubMedQA | **68.0** | 60.2 | 60.0 | 55.2 | 55.2 | 57.2 |
+| MedMCQA | **57.8** | 42.9 | 33.6 | 35.8 | 35.0 | 34.3 |
+| MedBullets | **42.9** | 26.0 | 20.1 | 20.5 | 19.2 | 17.2 |
+| MMLU (med) | **77.5** | 50.5 | 30.3 | 26.6 | 22.8 | 22.2 |
+| MMLU-Pro (med) | **47.3** | 20.7 | 13.6 | 13.6 | 14.4 | 9.4 |
+| MedExQA | **73.3** | 49.3 | 35.1 | 28.7 | 21.7 | 21.7 |
+| MedXpertQA-R | 12.1 | 10.5 | 9.6 | 11.4 | **18.1** | 11.0 |
+| MedXpertQA-U | 14.6 | 9.7 | 10.4 | 11.5 | **16.5** | 9.7 |
+| AfriMedQA | **48.9** | 32.2 | 25.3 | 12.6 | 12.6 | 10.9 |
+| **Macro avg** | **50.7** | 34.6 | 26.7 | 24.5 | 24.4 | 22.1 |
 
-Tanit-MedReason-8B was developed through a carefully designed 4-phase training pipeline:
+Tanit-Med-8B wins eight of ten. The two it loses are the two where nobody is really winning (see below).
 
-```
-┌─────────────────────────────────────────────────────────────────┐
-│  Phase 1: Broad Medical SFT (24.5 hours)                        │
-│  ├── 800K+ diverse medical samples                              │
-│  ├── Foundation knowledge acquisition                           │
-│  └── Loss: 1.129 → Comprehensive medical understanding          │
-├─────────────────────────────────────────────────────────────────┤
-│  Phase 2: Reasoning-Focused SFT (6.8 hours)                     │
-│  ├── 155K high-quality reasoning chains                         │
-│  ├── Focus: Medical-R1-Distill + Huatuo-o1                      │
-│  └── Loss: 1.014 → Enhanced reasoning capabilities              │
-├─────────────────────────────────────────────────────────────────┤
-│  Phase 3A: DPO Preference Alignment (5.9 hours)                 │
-│  ├── 33K preference pairs (FineMed-DPO)                         │
-│  ├── Direct Preference Optimization                             │
-│  └── Loss: 0.263 → Human-aligned responses                      │
-├─────────────────────────────────────────────────────────────────┤
-│  Phase 4: MedReason CoT Fine-Tuning (3 hours)                   │
-│  ├── 35K samples from UCSC-VLAA/MedReason                       │
-│  ├── Structured medical reasoning with <think> tags             │
-│  └── Loss: 1.662 → Production-ready model                       │
-└─────────────────────────────────────────────────────────────────┘
-```
+The AfriMedQA gap is the one we're proudest of: **48.9 vs 32.2** for the strongest baseline, and 4× the score of HuatuoGPT-o1 and MedReason. Medical models trained on US-exam corpora tend to fall over on questions grounded in African clinical practice. This one falls over less.
 
-**Total Training Time**: ~42 hours on NVIDIA H200 144GB
+### Against frontier models, for scale
 
+These are the published MedAgentsBench reference numbers ([arXiv:2503.07459](https://arxiv.org/abs/2503.07459)), **not re-run by us**. Cross-harness comparison, so read them as a ruler rather than a leaderboard. Averaged over the nine benchmarks the paper reports.
 
-## 🚀 Quick Start
+| Model | FULL (9-bench avg) | HARD (9-bench avg) |
+|---|---|---|
+| DeepSeek-R1 | 73.9 | 32.5 |
+| o3-mini | 71.8 | 28.0 |
+| GPT-4o | 68.2 | 18.0 |
+| o1-mini | 67.8 | 25.3 |
+| DeepSeek-V3 | 63.1 | 12.2 |
+| Claude-3.5-Sonnet | 61.8 | 12.3 |
+| Llama-3.3-70B | 61.8 | 12.4 |
+| QwQ-32B | 61.0 | 17.1 |
+| GPT-4o-mini | 57.8 | 10.8 |
+| Claude-3.5-Haiku | 55.0 | 12.0 |
+| **Tanit-Med-8B** | **50.9** | **20.4** |
 
-### Installation
+An 8B model running on a single consumer GPU lands about 5 points behind GPT-4o-mini and 4 behind Claude-3.5-Haiku on the standard splits. That is the honest position: competitive with last-generation *small* frontier models, not with frontier models.
 
-```bash
-pip install transformers torch accelerate
-```
+Please do **not** read the HARD column as "an 8B model beats GPT-4o." See the next section for why that number is a mirage.
 
-### Using with Transformers
+---
 
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-import torch
+## How to read the hard splits (please read this)
 
-model_name = "TanitAI/Tanit-MedReason-8B"
+The MedAgentsBench hard subsets were built by keeping only questions that **fewer than half of a panel of baseline LLMs answered correctly** — a panel that included GPT-4o. Those models are therefore *adversarially selected against* on this split. GPT-4o's 18.0 is not a measurement of GPT-4o being worse at medicine than an 8B model; it is a measurement of the filter having done its job.
 
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    torch_dtype=torch.bfloat16,
-    device_map="auto",
-    trust_remote_code=True
-)
+Three things follow, and we'd rather say them ourselves than have someone say them in the community tab:
 
-# Medical question
-question = """A 45-year-old male presents with sudden onset chest pain radiating to 
-the left arm, diaphoresis, and shortness of breath. ECG shows ST elevation in 
-leads V1-V4. What is the most likely diagnosis and immediate management?"""
+**1. At 8B, the hard splits do not separate models.** Here is every 8B-class model we tested, macro-averaged over the ten hard subsets:
 
-messages = [
-    {"role": "user", "content": question}
-]
+| Model | HARD macro avg |
+|---|---|
+| Falcon-H1R-7B | 21.5 |
+| MedReason-8B | 21.2 |
+| HuatuoGPT-o1-8B | 20.7 |
+| **Tanit-Med-8B** | **20.3** |
+| Ministral-3-8B-Reasoning | 19.6 |
+| DeepSeek-R1-0528-Qwen3-8B | 17.4 |
 
-# Apply chat template
-input_text = tokenizer.apply_chat_template(
-    messages, 
-    tokenize=False, 
-    add_generation_prompt=True
-)
+A 4-point spread across six models with wildly different training. Our +16-point advantage on the standard splits evaporates entirely. We do not claim a hard-split win, because there isn't one.
 
-inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
+**2. The samples are tiny.** AfriMedQA-hard is *32 questions* — one item is worth 3.1 points. Seven of the ten hard subsets are n=100. Treat any difference under ~8 points on these splits as noise, ours included.
 
-# Generate response
-with torch.no_grad():
-    outputs = model.generate(
-        **inputs,
-        max_new_tokens=2048,
-        temperature=0.6,
-        top_p=0.95,
-        do_sample=True
-    )
+**3. Bigger doesn't fix it, and self-consistency can make it worse.** In our own baseline sweep, Qwen3-30B-A3B with sampling + self-consistency (k=3) reached **60.2%** micro-average on the standard splits and **14.9%** on the hard ones — *below* Qwen3-0.6B's 20.2%. On adversarially-filtered items, majority voting appears to converge confidently on the attractive distractor. This was the most useful negative result of the project and it is why we ship a greedy, single-sample model.
 
-response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
-print(response)
-```
+---
 
-### Using with vLLM (Recommended for Production)
+## Where this model fails
 
-```python
-from vllm import LLM, SamplingParams
+- **MedXpertQA is unsolved at this scale.** Its items carry up to ten answer options, which puts chance around 10%. We score 12.1 (R) and 14.6 (U) — barely off the floor. So does every other 8B model here. DeepSeek-R1 gets 37.3. This is not a benchmark we are competitive on; it's a benchmark that shows where 8B runs out of road.
+- **MedReason-8B beats us on both MedXpertQA subsets** (18.1 / 16.5). Knowledge-graph-grounded training seems to buy something on deep-reasoning items that our curriculum doesn't.
+- **Hard splits: no better than the field.** See above.
+- **Long-form clinical advice is untested.** Every number in this card comes from multiple-choice benchmarks. We have not evaluated free-text safety, hedging, refusal behaviour, hallucinated citations, or drug dosing.
+- **English only.** AfriMedQA is English-language. We have run no non-English evaluation.
+- **Answer-format brittleness.** The phase-4 checkpoint is tuned to emit a strict final-answer line. Prompt it off-format and extraction gets flaky (see *Evaluation protocol*).
 
-model_name = "TanitAI/Tanit-MedReason-8B"
+### A note we owe you
 
-llm = LLM(
-    model=model_name,
-    dtype="bfloat16",
-    tensor_parallel_size=1,  # Adjust based on your GPU setup
-    trust_remote_code=True,
-    max_model_len=8192
-)
+We have not yet published a like-for-like **Qwen3-8B baseline** under this exact frozen harness. Our internal sweep of Qwen3-8B in thinking mode lands in a similar range to Tanit-Med-8B on several standard splits, and we are not going to make a claim of the form *"medical fine-tuning beats the base model"* until that number is measured under the same conditions as everything else in this card. Until it is up, read the peer table as a comparison against **other medical fine-tunes**, not as proof that medical SFT was worth it. That number is coming, and it will go here whatever it says.
 
-sampling_params = SamplingParams(
-    temperature=0.6,
-    top_p=0.95,
-    max_tokens=2048
-)
+---
 
-question = "What are the diagnostic criteria for Type 2 Diabetes Mellitus?"
+## Training
 
-# Format with chat template
-prompt = f"<|im_start|>user\n{question}<|im_end|>\n<|im_start|>assistant\n"
+Four stages, full fine-tune (no LoRA), ZeRO-3 / FSDP, bf16.
 
-outputs = llm.generate([prompt], sampling_params)
-print(outputs[0].outputs[0].text)
-```
+| Phase | Method | Data | Purpose |
+|---|---|---|---|
+| 1 | SFT | medical-o1-reasoning (Huatuo-o1) + Medical-R1-Distill + MedReason | Broad medical foundation, clinical language |
+| 2 | SFT | m23k + II-Medical-RL + ChatDoctor-RL + MedReason | MCQ robustness, multi-step reasoning structure |
+| 3 | DPO | FineMed-DPO (32.9K pairs) | Preference alignment; fewer confidently-wrong answers |
+| 4 | SFT | MedReason + Huatuo-o1, outputs rewritten to ≤6 reasoning bullets + final answer line | Concise, extractable chain-of-thought |
 
-### Using via API (Without Loading Model)
+### Hyperparameters
 
-For team members who need access without loading the model locally, use the HuggingFace Inference API:
+| | |
+|---|---|
+| Optimizer | AdamW, β = (0.9, 0.95), ε = 1e-8, weight decay 0.1 |
+| Grad clip | 1.0 |
+| Max sequence length | 8,192, packing on |
+| Attention | FlashAttention |
+| Global batch | ~1,048,576 tokens/step (≈128 × 8,192 seqs), grad-accum to match |
+| Schedule | 2% warmup, cosine decay to 10% of peak |
+| Peak LR — phase 1 | 2.0e-5 |
+| Peak LR — phase 2 | 1.0e-5 |
+| Peak LR — phase 3 (DPO) | 5.0e-7 |
+| Peak LR — phase 4 | 5.0e-6 |
 
-```python
-import requests
-
-API_URL = "https://api-inference.huggingface.co/models/TanitAI/Tanit-MedReason-8B"
-headers = {"Authorization": "Bearer YOUR_HF_TOKEN"}
-
-def query(payload):
-    response = requests.post(API_URL, headers=headers, json=payload)
-    return response.json()
-
-# Example query
-output = query({
-    "inputs": "What is the first-line treatment for hypertension?",
-    "parameters": {
-        "max_new_tokens": 1024,
-        "temperature": 0.6
-    }
-})
-print(output)
-```
+### What each phase actually bought
+
+Measured on our internal eval harness, macro-averaged across all ten standard splits:
 
-### Accessing via HuggingFace Hub
+- **Phase 1 → 2** was the single biggest jump. Reasoning SFT took MedBullets from 30.5 to 47.1 and MMLU-Pro from 25.7 to 43.8 — the compositional benchmarks, exactly where you'd hope forced multi-step reasoning would pay.
+- **Phase 3 (DPO)** *cost* raw accuracy. The DPO checkpoint sits several points below phase 4 on the standard splits and only reasons on 40–80% of prompts. What it bought was calibration: markedly fewer confident wrong answers. We ship it separately as [Tanit-Med-8B-DPO](https://huggingface.co/TanitAI/Tanit-Med-8B-DPO) because for some downstream uses that trade is the right one.
+- **Phase 4 (CoT polish)** recovered the accuracy *and* pushed the reasoning rate to ~99% of responses while shortening the traces. It also cut extraction failures by an order of magnitude, which — see below — turned out to matter more than we expected.
+
+We capped the reasoning budget at 4,096 tokens throughout. This follows [m1](https://arxiv.org/abs/2504.00869), which identifies an optimal medical-reasoning threshold around 4K tokens, past which accuracy *declines* — unlike in maths, forcing more reasoning on a medical question mostly gives a model with a shaky knowledge prior more rope to talk itself out of a correct answer. We saw the same thing and stopped fighting it.
+
+---
+
+## Usage
 
 ```python
-from huggingface_hub import InferenceClient
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-client = InferenceClient(
-    model="TanitAI/Tanit-MedReason-8B",
-    token="YOUR_HF_TOKEN"
-)
+model_id = "TanitAI/Tanit-Med-8B"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto", device_map="auto")
 
-response = client.text_generation(
-    "Explain the pathophysiology of heart failure.",
-    max_new_tokens=1024,
-    temperature=0.6
+messages = [
+    {"role": "system", "content": "You are Tanit, an expert medical AI assistant."},
+    {"role": "user", "content": (
+        "A 45-year-old presents with acute monoarthritis of the first MTP joint.\n"
+        "What is the most likely diagnosis?\n\n"
+        "A. Gout\nB. Pseudogout\nC. Septic arthritis\nD. Rheumatoid arthritis"
+    )},
+]
+
+prompt = tokenizer.apply_chat_template(
+    messages, tokenize=False, add_generation_prompt=True, enable_thinking=True
 )
-print(response)
+inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+out = model.generate(**inputs, max_new_tokens=4096, temperature=0.6, top_p=0.95, top_k=20)
+print(tokenizer.decode(out[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True))
 ```
 
-## 📊 Evaluation Results
+**Serving with vLLM:**
 
-| Benchmark | Accuracy |
-|-----------|----------|
-| MedQA | 60.3% |
-| MMLU Medical | 66.4% |
-| MedExQA | 62.1% |
-| PubMedQA | 74.6% |
-| MedMCQA | 52.9% |
+```bash
+vllm serve TanitAI/Tanit-Med-8B --max-model-len 32768 --reasoning-parser qwen3
+```
 
-*Evaluated using zero-shot prompting with chain-of-thought reasoning.*
+**Generation settings.** For open-ended use, follow Qwen3's thinking-mode defaults: `temperature=0.6`, `top_p=0.95`, `top_k=20`. For benchmark reproduction, use **greedy** (`temperature=0`) — every number in this card was produced that way. Do not use greedy decoding with `enable_thinking=True` for long open-ended generations; it degenerates into repetition, as it does for the base model.
 
+**Quantized:** [Tanit-Med-8B-NVFP4](https://huggingface.co/TanitAI/Tanit-Med-8B-NVFP4) runs on a single consumer GPU.
 
-## 📋 Training Details
+---
+
+## Evaluation protocol
+
+Reproducing our numbers requires matching three things, in descending order of how much they'll bite you:
+
+1. **Answer extraction.** This is a bigger confound than most training decisions. Between two revisions of our own harness, the strict-format extraction-failure rate on this checkpoint moved from **31.3% → 4.8%** on MedQA (and 63% → 5% on MedQA-hard), while accuracy moved by about a point. If you benchmark against us and get a different number, check your extraction rate *first*. Ours is under 7% on every split.
+2. **Prompt.** We use the MedAgentsBench zero-shot prompt verbatim — knowledgeable-medical-assistant system message, question, options, "reply with the answer index only."
+3. **Decoding.** Greedy. Single sample. Reasoning budget 4,096 tokens, generation cap 16,384.
+
+Self-consistency (k=5) is worth roughly **+3–10% relative** on the standard splits if you can afford 5× the compute — but re-read the hard-split warning above before you assume it helps everywhere. It doesn't.
 
-| Parameter | Value |
-|-----------|-------|
-| **Base Model** | `deepseek-ai/DeepSeek-R1-0528-Qwen3-8B` |
-| **Training Phase** | Final (v1.0) |
-| **Training Data** | 35K samples from UCSC-VLAA/MedReason dataset with structured medical reasoning |
-| **Training Steps** | 4,442 |
-| **Training Time** | 3.0 hours |
-| **Final Loss** | 1.662 |
-| **Precision** | bfloat16 |
-| **Context Length** | 8,192 tokens |
+---
 
-## ⚠️ Limitations & Intended Use
+## The collection
 
-### Intended Use
-- Medical education and research
-- Clinical decision support (with physician oversight)
-- Medical question answering
-- Healthcare documentation assistance
+| Model | Stage | Use it if |
+|---|---|---|
+| [**Tanit-Med-8B**](https://huggingface.co/TanitAI/Tanit-Med-8B) | Phase 4 (CoT polish) | You want the best accuracy. **Start here.** |
+| [Tanit-Med-8B-DPO](https://huggingface.co/TanitAI/Tanit-Med-8B-DPO) | Phase 3 (DPO) | You'd trade a few points of accuracy for better-calibrated confidence |
+| [Tanit-MedReason-8B](https://huggingface.co/TanitAI/Tanit-MedReason-8B) | Phase 2 (reasoning SFT) | You want an un-DPO'd reasoning checkpoint to build on |
+| [Tanit-Med-8B-NVFP4](https://huggingface.co/TanitAI/Tanit-Med-8B-NVFP4) | Phase 4, FP4 | You're deploying on a consumer GPU |
 
-### Limitations
-- **Not a replacement for professional medical advice**
-- May generate plausible-sounding but incorrect information
-- Performance varies across medical specialties
-- Should always be used with human oversight in clinical settings
+---
 
-### Ethical Considerations
-- This model is intended to assist, not replace, healthcare professionals
-- Always verify medical information with authoritative sources
-- Do not use for diagnosis or treatment without professional consultation
+## Intended use and limitations
 
-## 📜 License
+**Intended for:** medical NLP research, benchmark development, medical education tooling, retrieval-augmented clinical QA prototypes, and as a base for further fine-tuning.
 
-This model is released under the [Apache 2.0 License](https://opensource.org/licenses/Apache-2.0).
+**Not intended for, and not safe for:** clinical decision support, diagnosis, treatment or dosing recommendations, triage, or any patient-facing deployment. Tanit-Med-8B is **not a medical device**. It has not been reviewed or cleared by any regulator, has not been evaluated for clinical safety, and has been measured only on multiple-choice exam questions — which correlate with medical knowledge but are not a proxy for clinical judgment. It answers with a `<think>` trace that *looks* like reasoning; a confident, fluent, well-structured trace can and does precede a wrong answer.
 
-## 🙏 Acknowledgments
+**Known risks:** inherits the biases of its training corpora, which are overwhelmingly Western and exam-derived. Will hallucinate drug names, doses, and citations. Will not reliably refuse out-of-scope or unsafe requests — DPO here optimized for *correctness* under ambiguity, not for safety refusals.
 
-- Built on [DeepSeek-R1-0528-Qwen3-8B](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B)
-- Training data from FreedomIntelligence, UCSC-VLAA, Intelligent-Internet, and other open-source medical datasets
-- Developed by [TANIT Healthcare Technologies](https://tanitai.com)
+If you deploy anything downstream of this model in a setting where a person could be harmed, that safety work is yours to do, and it has not been done here.
 
-## 📧 Contact
+---
 
-For questions, collaborations, or access requests:
-- **Organization**: [TANIT Healthcare Technologies](https://huggingface.co/TanitAI)
-- **Email**: contact@tanitai.com
+## Licensing
+
+Model weights are released under **Apache-2.0**, inherited from Qwen3-8B. Note that the training corpora carry their own terms — several are research-oriented and some derive from sources with non-commercial restrictions. If you intend to use this model commercially, verify the licence of each dataset listed in the metadata; we are releasing weights, not indemnity.
 
 ---
 
-<div align="center">
+## Citation
 
-**Made with ❤️ by TANIT Healthcare Technologies**
+```bibtex
+@misc{tanit-med-8b-2026,
+  title  = {Tanit-Med-8B: A Four-Stage Curriculum for Open Medical Reasoning at 8B},
+  author = {Tanit Healthcare Technologies},
+  year   = {2026},
+  url    = {https://huggingface.co/TanitAI/Tanit-Med-8B}
+}
+```
 
-*Advancing healthcare through AI*
+Built on [Qwen3](https://huggingface.co/Qwen/Qwen3-8B) (Qwen Team). Evaluated with [MedAgentsBench](https://arxiv.org/abs/2503.07459) (Zhang et al., 2025). Trained on data from [FreedomIntelligence](https://huggingface.co/FreedomIntelligence), [UCSC-VLAA](https://huggingface.co/UCSC-VLAA), [Intelligent-Internet](https://huggingface.co/Intelligent-Internet), and [FineMed](https://huggingface.co/hongzhouyu). Our thanks to all of them — the open medical-AI stack is a shared one.
 
-</div>
+**Found a problem with these numbers?** Open a discussion. We'd rather be corrected than cited wrongly.