clarity-ai 6.5.2 → 6.5.4

package/CHANGELOG.md

@@ -2,6 +2,16 @@
 
 ---
 
+## 6.5.4 (2026-06-06)
+
+### Docker SDK + Local GGUF Inference
+- All 6 Spaces converted to `sdk: docker` with FastAPI + `llama-cpp-python`
+- Models run locally on each Space (no HF Inference API dependency)
+- Sync FastAPI handlers for compatibility; lazy model load on first request
+- Pre-built CPU wheels for `llama-cpp-python` via `--extra-index-url`
+- Spaces made public — no auth header needed
+- Default model: Qwen2.5-1.5B-Instruct-Q4_K_M (Flash & Heavy)
+
 ## 6.5.2 (2026-06-06)
 
 ### CoT + MoE Space Architecture

package/Dockerfile

@@ -0,0 +1,9 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY app.py .
+
+EXPOSE 7860
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "clarity-ai",
-  "version": "6.5.2",
-  "description": "CLARITY — terminal AI agent with Clarity Flash 14B",
+  "version": "6.5.4",
+  "description": "CLARITY — terminal AI agent with local GGUF inference on HF Spaces",
   "type": "module",
   "bin": {
     "clarity": "bin/clarity.js"

package/requirements.txt

@@ -1,9 +1,4 @@
+gradio>=5
+huggingface-hub>=0.22
+httpx>=0.27
 fastapi>=0.100
-uvicorn>=0.20
-transformers>=4.38
-torch>=2.1
-accelerate>=0.27
-bitsandbytes>=0.42
-peft>=0.8
-huggingface-hub>=0.20
-sentencepiece>=0.1.99

package/space_app.py

@@ -1,83 +1,26 @@
-import os, json, time, asyncio
-from threading import Thread
-from fastapi import FastAPI, Request
-from fastapi.responses import StreamingResponse, JSONResponse
-from contextlib import asynccontextmanager
+import gradio as gr
+from huggingface_hub import InferenceClient
+import os
 
 HF_TOKEN = os.environ.get('HF_TOKEN', '')
 MODEL_NAME = os.environ.get('HF_MODEL', 'Clarity-flash-weights')
-BASE_MODEL = os.environ.get('BASE_MODEL_ID', '')
-DS = f'Universal-618/{MODEL_NAME}'
 
-print(f'[startup] model={MODEL_NAME} base={BASE_MODEL}', flush=True)
-
-model = None
-tokenizer = None
-
-def load_model():
-    global model, tokenizer
-    print('[startup] downloading weights...', flush=True)
-    from huggingface_hub import snapshot_download
-    path = snapshot_download(repo_id=DS, repo_type='dataset', token=HF_TOKEN)
-    print(f'[startup] weights at {path}', flush=True)
-    from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig, TextIteratorStreamer
-    from peft import PeftModel
-    bnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype='float16', bnb_4bit_use_double_quant=True)
-    print('[startup] loading base model...', flush=True)
-    base = AutoModelForCausalLM.from_pretrained(
-        BASE_MODEL, quantization_config=bnb, device_map='auto',
-        trust_remote_code=True, low_cpu_mem_usage=True, token=HF_TOKEN)
-    print('[startup] loading LoRA...', flush=True)
-    model = PeftModel.from_pretrained(base, path, token=HF_TOKEN)
-    tokenizer = AutoTokenizer.from_pretrained(path, token=HF_TOKEN)
-    if tokenizer.pad_token is None:
-        tokenizer.pad_token = tokenizer.eos_token
-    print('[startup] ready', flush=True)
-
-load_model()
-
-app = FastAPI()
-
-@app.post('/v1/chat/completions')
-async def generate(request: Request):
-    body = await request.json()
-    prompt = tokenizer.apply_chat_template(body['messages'], tokenize=False, add_generation_prompt=True)
-    inputs = tokenizer(prompt, return_tensors='pt')
-    stream = body.get('stream', True)
-    max_tokens = body.get('max_tokens', 4096)
-    gen_kwargs = dict(input_ids=inputs.input_ids, attention_mask=inputs.attention_mask,
-                      max_new_tokens=max_tokens, temperature=body.get('temperature', 0.7),
-                      do_sample=True, pad_token_id=tokenizer.pad_token_id)
-    if stream:
-        from transformers import TextIteratorStreamer
-        streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
-        t = Thread(target=model.generate, kwargs={**gen_kwargs, 'streamer': streamer})
-        t.start()
-        async def gen():
-            yield 'data: {"choices":[{"delta":{"role":"assistant"},"index":0}]}\n\n'
-            for text in streamer:
-                if text:
-                    yield f'data: {json.dumps({"choices":[{"delta":{"content":text},"index":0}]})}\n\n'
-            yield 'data: [DONE]\n\n'
-        return StreamingResponse(gen(), media_type='text/event-stream', headers={
-            'Cache-Control': 'no-cache', 'Connection': 'keep-alive', 'X-Accel-Buffering': 'no'})
-    out = model.generate(**gen_kwargs)
-    text = tokenizer.decode(out[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
-    return {'choices': [{'message': {'role': 'assistant', 'content': text}, 'index': 0}]}
-
-@app.get('/main-data/{path:path}')
-async def main_data(path: str):
-    path = path or 'index.json'
-    fp = f'/data/{path}'
-    if not os.path.exists(fp):
-        return JSONResponse({'error': 'not found'}, status_code=404)
-    with open(fp) as f:
-        return JSONResponse(json.load(f))
-
-@app.get('/')
-async def root():
-    return {'status': 'ok', 'model': MODEL_NAME}
-
-if __name__ == '__main__':
-    import uvicorn
-    uvicorn.run(app, host='0.0.0.0', port=7860)
+def respond(message, history, system_message, max_tokens, temperature, top_p):
+    client = InferenceClient(token=HF_TOKEN, model=f'Universal-618/{MODEL_NAME}')
+    messages = [{"role": "system", "content": system_message}] + history + [{"role": "user", "content": message}]
+    response = ''
+    for chunk in client.chat_completion(messages, max_tokens=max_tokens, stream=True, temperature=temperature, top_p=top_p):
+        if chunk.choices and chunk.choices[0].delta.content:
+            response += chunk.choices[0].delta.content
+            yield response
+
+gr.ChatInterface(
+    respond,
+    additional_inputs=[
+        gr.Textbox(value="You are a helpful assistant.", label="System message"),
+        gr.Slider(minimum=1, maximum=8192, value=4096, step=1, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
+    ],
+    title=f'CLARITY {MODEL_NAME}',
+)

package/src/config/models.js

@@ -1,6 +1,6 @@
 export const ALL_MODELS = [
-  { id: 'huggingface/Universal-618/Clarity-flash-weights', provider: 'huggingface', label: 'Clarity Flash 14B', badge: '128K' },
-  { id: 'huggingface/Universal-618/Clarity-heavy-weights', provider: 'huggingface', label: 'Clarity Heavy 20B', badge: 'MoE' },
+  { id: 'huggingface/Universal-618/Clarity-flash-weights', provider: 'huggingface', label: 'Clarity Flash (Qwen 2.5 1.5B)', badge: 'Free' },
+  { id: 'huggingface/Universal-618/Clarity-heavy-weights', provider: 'huggingface', label: 'Clarity Heavy (Qwen 2.5 1.5B)', badge: 'Free' },
   { id: 'groq/llama-3.3-70b-versatile', provider: 'groq', label: 'Llama 3.3 70B Versatile', badge: null },
   { id: 'groq/llama-3.1-8b-instant', provider: 'groq', label: 'Llama 3.1 8B Instant', badge: 'Fast' },
   { id: 'groq/llama-4-scout-17b-16e-instruct', provider: 'groq', label: 'Llama 4 Scout 17B', badge: null },

package/src/providers/index.js

@@ -3,12 +3,12 @@ import { streamResponse } from './streaming.js';
 
 const MODEL_ROUTES = {
   'Universal-618/Clarity-flash-weights': [
+    process.env.CLARITY_INFERENCE_URL || 'https://universal-618-clarity-main.hf.space/v1/chat/completions',
     'https://universal-618-clarity-2.hf.space/v1/chat/completions',
-    'https://universal-618-clarity-4.hf.space/v1/chat/completions',
-    'https://universal-618-clarity-5.hf.space/v1/chat/completions',
+    'https://universal-618-clarity-3.hf.space/v1/chat/completions',
   ],
   'Universal-618/Clarity-heavy-weights': [
-    'https://universal-618-clarity-3.hf.space/v1/chat/completions',
+    process.env.CLARITY_INFERENCE_URL || 'https://universal-618-clarity-4.hf.space/v1/chat/completions',
     'https://universal-618-clarity-5.hf.space/v1/chat/completions',
     'https://universal-618-clarity-6.hf.space/v1/chat/completions',
   ],

package/src/providers/streaming.js

@@ -26,7 +26,7 @@ export async function* streamResponse(endpoint, body, apiKey, extraHeaders = {},
       method: 'POST',
       headers: {
         'Content-Type': 'application/json',
-        'Authorization': 'Bearer ' + apiKey,
+        ...(apiKey && apiKey !== 'none' ? { 'Authorization': 'Bearer ' + apiKey } : {}),
         ...extraHeaders,
       },
       body: JSON.stringify({ ...body, stream: true }),