feat: update acne assistant lite gguf 0.6b

.gitattributes

@@ -33,3 +33,11 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+gguf/airy-0.6b-iq1_m.gguf filter=lfs diff=lfs merge=lfs -text
+gguf/airy-0.6b-iq1_s.gguf filter=lfs diff=lfs merge=lfs -text
+gguf/airy-0.6b-iq2_m.gguf filter=lfs diff=lfs merge=lfs -text
+gguf/airy-0.6b-iq2_xs.gguf filter=lfs diff=lfs merge=lfs -text
+gguf/airy-0.6b-iq2_xxs.gguf filter=lfs diff=lfs merge=lfs -text
+gguf/airy-0.6b-iq3_m.gguf filter=lfs diff=lfs merge=lfs -text
+gguf/airy-0.6b-q2_k.gguf filter=lfs diff=lfs merge=lfs -text
+gguf/airy-0.6b-q3_k_m.gguf filter=lfs diff=lfs merge=lfs -text

PROMPT_TEMPLATE.txt

@@ -0,0 +1,36 @@
+System prompt (default, injected automatically if no system message provided):
+
+You are Acnoryx AI, a friendly dermatology assistant in the Acnoryx acne scanner app.
+Your expertise: acne types, causes, treatments, skincare routines.
+Only answer skincare/dermatology topics. Politely decline unrelated questions.
+You are NOT a doctor. Be clear, friendly, and helpful.
+Use markdown: ## headings, - bullets, **bold** for key terms, icons for visual clarity.
+For medical/skin analysis topics, always add a disclaimer note.
+Keep the answer clear, useful, and moderately detailed.
+Reason carefully before answering.
+Put your reasoning inside <think>...</think>, then provide the final user-facing answer after the closing tag.
+
+---
+
+Thinking mode (recommended):
+  A prompt tag <think> triggers internal reasoning before the answer.
+  Compatible runtimes handle this automatically via the chat template.
+
+Non-thinking mode:
+  Append /no_think to the system prompt, or use an explicit user instruction.
+
+---
+
+Example prompts (Vietnamese):
+- Xin chào
+- Bạn là ai?
+- Da em nhiều mụn ẩn ở trán thì nên bắt đầu từ đâu?
+- Kết quả quét của tôi: mụn cám (28%), mụn nang (19%), mụn mủ (35%). Hãy tóm tắt.
+- Phân tích AI vùng má phải: mụn mủ (41%), thâm đỏ sau mụn (24%), mụn cám (20%), mụn nang (15%).
+
+Example prompts (English):
+- Hello, who are you?
+- I have inflamed acne and dark marks. What should I prioritize?
+- What is retinol and can I use it for acne?
+- My scan result: comedones (28%), cysts (19%), pustules (35%). Please summarize.
+- Explain the difference between blackheads and whiteheads.

README.md

@@ -0,0 +1,190 @@
+---
+language:
+  - vi
+  - en
+license: other
+base_model:
+  - Acnoryx/Airy
+  - Qwen/Qwen3-0.6B
+library_name: gguf
+pipeline_tag: text-generation
+tags:
+  - acne
+  - skincare
+  - dermatology
+  - gguf
+  - llama-cpp
+  - fine-tuned
+  - airy
+  - acnoryx
+quantized_by: Acnoryx
+inference: false
+---
+
+# Acnoryx/Airy-Lite
+
+**Airy-Lite** is the ultra-low-bit, ultra-lightweight research branch of the Acnoryx model family.
+
+This repository contains GGUF quantizations of the same fine-tuned weights used in the Acnoryx skincare model  
+(Qwen/Qwen3-0.6B base → Acnoryx/Airy → Airy-Lite), optimized for ultra-constrained environments and research.
+
+---
+
+## Overview
+
+Both model branches share the same fine-tuned base weights, but differ in quantization depth:
+
+| Branch | Quantization range | Target |
+|--------|--------------------|--------|
+| **Acnoryx-Core** (release) | Q4_0 → F16 | Production / deployment |
+| **Airy-Lite** (this repo) | Q3_K_M → IQ1_S | Research / ultra-constrained edge |
+
+> **Note:** Airy-Lite is provided for research and evaluation purposes only.  
+> The Acnoryx app uses Acnoryx-Core variants, not Airy-Lite.  
+> See the live app on Google Play for production usage.
+
+---
+
+## Evaluation Against Acnoryx-Core
+
+**Test profile:** `full-100` — 100 questions × 2 modes (thinking / non-thinking)  
+**Categories:** acne knowledge, app facts, refusal, skincare, identity, scan analysis, reasoning, medical uncertainty.
+
+### Acnoryx-Core (Reference)
+
+| Model | Size | Think % | No-Think % | Rating |
+|-------|------|---------|-----------|--------|
+| acnoryx-0.6b-f16 | 1143 MB | 99.0% | 100.0% | ✅ Excellent |
+| acnoryx-0.6b-q8_0 | 610 MB | 98.0% | 99.0% | ✅ Excellent |
+| acnoryx-0.6b-q5_k_m | 424 MB | 98.0% | 96.0% | ✅ Excellent |
+| acnoryx-0.6b-q4_k_m | 378 MB | 99.0% | 99.0% | ✅ Excellent |
+| acnoryx-0.6b-q4_0 | 364 MB | 94.0% | 96.0% | ✅ Good |
+| acnoryx-0.6b-iq4_nl | 365 MB | 96.0% | 97.0% | ✅ Good |
+| acnoryx-0.6b-iq4_xs | 352 MB | 96.0% | 94.0% | ✅ Good |
+
+### Airy-Lite (This Repository)
+
+| Model | Size | Think % | No-Think % | Rating |
+|-------|------|---------|-----------|--------|
+| airy-0.6b-q3_k_m | 331 MB | 92.0% | 85.0% | ⚠️ Degraded |
+| airy-0.6b-iq3_m | 320 MB | 18.0% | 17.0% | ❌ Collapsed |
+| airy-0.6b-q2_k | 282 MB | 17.0% | 11.0% | ❌ Collapsed |
+| airy-0.6b-iq2_m | 253 MB | 14.0% | 8.0% | ❌ Collapsed |
+| airy-0.6b-iq2_xs | 231 MB | 0.0% | 4.0% | ❌ Collapsed |
+| airy-0.6b-iq2_xxs | 219 MB | 4.0% | 4.0% | ❌ Collapsed |
+| airy-0.6b-iq1_m | 206 MB | 3.0% | 3.0% | ❌ Collapsed |
+| airy-0.6b-iq1_s | 198 MB | 4.0% | 3.0% | ❌ Collapsed |
+
+---
+
+## Verdict
+
+A sharp quality drop occurs between:
+
+- **Q3_K_M** → still usable (degraded but coherent)
+- **IQ3_M and below** → completely collapsed
+
+All IQ2 and IQ1 variants produce incoherent output regardless of format.
+
+👉 **Only `airy-0.6b-q3_k_m` is usable** for experimentation.  
+👉 All other variants are strictly for research/benchmarking.
+
+**For production: use Acnoryx-Core (Q4_0 or higher).**
+
+---
+
+## Usability Classification
+
+| Level | Models | Status |
+|-------|--------|--------|
+| ⚠️ Usable | airy-0.6b-q3_k_m | Degraded but coherent |
+| ❌ Not usable | All IQ3, IQ2, IQ1 variants | Garbled / incoherent output |
+
+> Do not deploy Airy-Lite models in production systems.
+
+---
+
+## Files
+
+```text
+gguf/
+  airy-0.6b-q3_k_m.gguf     331 MB   ⚠️ Usable
+  airy-0.6b-iq3_m.gguf      320 MB   ❌ Collapsed
+  airy-0.6b-q2_k.gguf       282 MB   ❌ Collapsed
+  airy-0.6b-iq2_m.gguf      253 MB   ❌ Collapsed
+  airy-0.6b-iq2_xs.gguf     231 MB   ❌ Collapsed
+  airy-0.6b-iq2_xxs.gguf    219 MB   ❌ Collapsed
+  airy-0.6b-iq1_m.gguf      206 MB   ❌ Collapsed
+  airy-0.6b-iq1_s.gguf      198 MB   ❌ Collapsed
+
+PROMPT_TEMPLATE.txt
+README.md
+```
+
+---
+
+## How to Run
+
+### llama.cpp (CLI)
+
+```bash
+./llama-cli -m ./gguf/airy-0.6b-q3_k_m.gguf -cnv -p "Bạn là ai?"
+```
+
+---
+
+### Python (llama-cpp-python)
+
+```python
+from llama_cpp import Llama
+
+model = Llama(
+    model_path="./gguf/airy-0.6b-q3_k_m.gguf",
+    n_ctx=2048,
+)
+
+SYSTEM_PROMPT = (
+    "You are Acnoryx AI, a dermatology assistant. "
+    "Focus only on acne and skincare. "
+    "Politely refuse unrelated topics. "
+    "You are not a doctor."
+)
+
+response = model.create_chat_completion(
+    messages=[
+        {"role": "system", "content": SYSTEM_PROMPT},
+        {"role": "user", "content": "Mụn đầu đen là gì?"},
+    ],
+)
+
+print(response["choices"][0]["message"]["content"])
+```
+
+---
+
+## Intended Use
+
+- Research on ultra-low-bit quantization
+- Edge AI experimentation
+- GGUF benchmarking
+
+---
+
+## Limitations
+
+- Severe quality degradation below Q3
+- Not reliable for medical guidance
+- Not suitable for production deployment
+
+---
+
+## About Acnoryx
+
+Acnoryx is an AI-powered acne scanner app on Android.
+
+- Author: Nguyễn Trung Nhẫn  
+- Publisher: 5CANH APPS  
+- Contact: hello@5canh.com  
+
+Production uses **Acnoryx-Core**, not Airy-Lite.  
+Airy-Lite is released for research and community experimentation.

chat_template.jinja

@@ -0,0 +1,29 @@
+{%- set default_system = 'You are Acnoryx AI, a friendly dermatology assistant in the Acnoryx acne scanner app. Your expertise: acne types, causes, treatments, skincare routines. Only answer skincare/dermatology topics. Politely decline unrelated questions. You are NOT a doctor. Be clear, friendly, and helpful. Use markdown: ## headings, - bullets, **bold** for key terms, icons for visual clarity. For medical/skin analysis topics, always add a disclaimer note. Keep the answer clear, useful, and moderately detailed. Reason carefully before answering. Put your reasoning inside <think>...</think>, then provide the final user-facing answer after the closing tag.' -%}
+{%- if messages|length == 0 or messages[0].role != 'system' -%}
+{{- '<|im_start|>system
+' + default_system + '<|im_end|>
+' -}}
+{%- endif -%}
+{%- for message in messages -%}
+{%- if message.role == 'assistant' and '</think>' in message.content -%}
+{%- set parts = message.content.split('</think>') -%}
+{%- set thinking = parts[0].replace('<think>', '').strip() -%}
+{%- set answer = parts[1].strip() -%}
+{{- '<|im_start|>assistant
+<think>
+' + thinking + '
+</think>
+
+' + answer + '<|im_end|>
+' -}}
+{%- else -%}
+{{- '<|im_start|>' + message.role + '
+' + message.content + '<|im_end|>
+' -}}
+{%- endif -%}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+{{- '<|im_start|>assistant
+<think>
+' -}}
+{%- endif -%}

gguf/airy-0.6b-iq1_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efdef51aca25203c856615f1c8757cf17c026c068c307f2595bb75d57787b704
+size 216048768

gguf/airy-0.6b-iq1_s.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:855a20c2bf96b70cf0be3ded72ce59934229c97ee29843bf4d31de6bace7c902
+size 208012416

gguf/airy-0.6b-iq2_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10063e2736a96610c8bfa00090c0b2fb29cd711688d77f752dc607ce2ec77df7
+size 264905856

gguf/airy-0.6b-iq2_xs.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:abfd9a34b52b501235164c7ddcaf482c2420a762a5dbf95a76f8e65fc94a7564
+size 241992832

gguf/airy-0.6b-iq2_xxs.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e25c62aaa5a405a476117f974c3c1cf65966cf455c82107e9a424e89a4c75f5
+size 229442688

gguf/airy-0.6b-iq3_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e8803ee711059942c18d186362966f31b2f73a47f588c76017ee4e6e0d9ed60
+size 336023488

gguf/airy-0.6b-q2_k.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ace9d66fe60d31cba554d8f58e6ac1e9cddd85fa54741ae297948da994d0de5d
+size 296234944

gguf/airy-0.6b-q3_k_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43a21953b4def20cbfb8c8967ce54e60fecb14dfa15ca8e93fc07d72c6a653a5
+size 347123648