Text Generation
Safetensors
GGUF
Chinese
diet-nutrition
patient-education
qa
rag
qwen3
lora
retrieval-augmented-generation
health-routing
conversational
Instructions to use chenhaodev/diet-nutrition-qa with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use chenhaodev/diet-nutrition-qa with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf chenhaodev/diet-nutrition-qa:Q8_0 # Run inference directly in the terminal: llama cli -hf chenhaodev/diet-nutrition-qa:Q8_0
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf chenhaodev/diet-nutrition-qa:Q8_0 # Run inference directly in the terminal: llama cli -hf chenhaodev/diet-nutrition-qa:Q8_0
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf chenhaodev/diet-nutrition-qa:Q8_0 # Run inference directly in the terminal: ./build/bin/llama cli -hf chenhaodev/diet-nutrition-qa:Q8_0
Use Docker
docker model run hf.co/chenhaodev/diet-nutrition-qa:Q8_0
- Llama new
- LM Studio
- Jan
- vLLM
How to use chenhaodev/diet-nutrition-qa with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "chenhaodev/diet-nutrition-qa" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "chenhaodev/diet-nutrition-qa", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/chenhaodev/diet-nutrition-qa:Q8_0
- Ollama
How to use chenhaodev/diet-nutrition-qa with Ollama:
ollama run hf.co/chenhaodev/diet-nutrition-qa:Q8_0
- Unsloth Desktop
- Pi
How to use chenhaodev/diet-nutrition-qa with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf chenhaodev/diet-nutrition-qa:Q8_0
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:9931/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "chenhaodev/diet-nutrition-qa:Q8_0" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use chenhaodev/diet-nutrition-qa with Docker Model Runner:
docker model run hf.co/chenhaodev/diet-nutrition-qa:Q8_0
- Lemonade
How to use chenhaodev/diet-nutrition-qa with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull chenhaodev/diet-nutrition-qa:Q8_0
Run and chat with the model
lemonade run user.diet-nutrition-qa-Q8_0
List all available models
lemonade list
- Hermes Agent
How to use chenhaodev/diet-nutrition-qa with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf chenhaodev/diet-nutrition-qa:Q8_0
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:9931/v1 hermes config set model.default chenhaodev/diet-nutrition-qa:Q8_0
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use chenhaodev/diet-nutrition-qa with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf chenhaodev/diet-nutrition-qa:Q8_0
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:9931/v1 \ --custom-model-id "chenhaodev/diet-nutrition-qa:Q8_0" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
File size: 11,892 Bytes
709d759 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 | #!/usr/bin/env python3
"""Build router (small model) SFT dataset for the diet/nutrition Q&A harness.
Sources (weak supervision, consistent with the rule engine):
(1) REAL curated questions from the 3 books -> category = book chapter (authoritative),
intent from a keyword classifier, mode/red_flags from rules.
(2) AUGMENTED alerts : chronicity / acuteness / progression / domain red-flag entities
(GI bleed, diabetes hypo/DKA, osteoporosis fall/fracture).
(3) AUGMENTED intents : what-eat / what-avoid / diet-advice / nutrient / meal-plan /
lifestyle / definition / when-seek-care coverage.
(4) AUGMENTED compound: parallel multi-sub-intent queries.
Every example passes through scripts.red_flag_rules.scan so the supervised label is
EXACTLY what the harness will produce at inference (train/serve consistency).
For real curated questions the category is overridden to the authoritative book value.
Output: data/router/router_sft_train.jsonl + router_sft_val.jsonl
"""
import json
import os
import random
import sys
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from scripts.red_flag_rules import scan
ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
OUT = os.path.join(ROOT, "data/router")
os.makedirs(OUT, exist_ok=True)
# intent keyword classifier used to label real QA questions consistently
INTENT_KEYS = [
("what-avoid", ["禁忌", "不能吃", "不宜", "忌", "避免", "远离", "不要吃", "禁", "少吃", "不要喝"]),
("what-eat", ["吃多少", "可以吃", "能吃", "吃什么", "吃合适", "吃好", "吃高", "选择肉类", "挑选", "多吃", "吃些", "吃食", "吃哪些"]),
("meal-plan", ["食谱", "一日", "三餐", "早餐", "午餐", "晚餐", "搭配", "餐次", "加餐", "一周", "烹调", "烹饪", "做法"]),
("nutrient", ["补钙", "钙", "维生素", "营养", "蛋白质", "膳食纤维", "矿物质", "营养素", "脂肪", "盐", "糖", "含钙", "铁", "锌", "磷", "镁", "氨基酸"]),
("lifestyle", ["运动", "锻炼", "晒太阳", "阳光", "作息", "睡眠", "戒烟", "戒酒", "饮酒", "抽烟", "生活习惯", "心情"]),
("when-seek-care", ["就医", "看医生", "医院", "就诊", "医生", "检测", "检查", "诊断", "胃镜", "肠镜"]),
("definition", ["什么是", "是什么", "认识", "了解", "怎么回事", "什么样"]),
("diet-advice", ["饮食", "膳食", "调理", "营养治疗", "怎么吃", "护理"]),
]
def rule_intent(text):
for it, kws in INTENT_KEYS:
if any(k in text for k in kws):
return it
return "other"
def build_example(utterance, plan):
return {
"messages": [
{"role": "user",
"content": f"请把下面这句话拆成结构化的患者饮食意图计划。只输出 JSON。\n患者:{utterance}"},
{"role": "assistant", "content": json.dumps(plan, ensure_ascii=False)},
],
"utterance": utterance,
}
def plan_of(utterance, override_cat=None, override_intent=None):
red, subs, mode = scan(utterance)
plan = {"mode": mode, "red_flags": red if mode == "alert" else [], "sub_intents": subs}
if subs and (override_cat or override_intent):
s = plan["sub_intents"][0]
if override_cat:
s["category"] = override_cat
if override_intent:
s["intent"] = override_intent
return plan
# ========================= 1) REAL curated questions =========================
def real_questions():
recs = []
for key in ("gastro", "diabetes", "osteoporosis"):
p = os.path.join(ROOT, "data", f"qa_{key}.jsonl")
for line in open(p, encoding="utf-8"):
r = json.loads(line)
u = r["question"].rstrip("。").rstrip("?") + "?"
cat = r["category"]
intent = rule_intent(u)
plan = plan_of(u, override_cat=cat, override_intent=intent)
recs.append(build_example(u, plan))
return recs
# ========================= 2) AUGMENTED alerts =========================
CAT_ENTITY = {
"gastritis": ["胃痛", "胃胀", "反酸", "胃病反复", "浅表性胃炎"],
"peptic-ulcer": ["胃溃疡", "十二指肠溃疡", "溃疡又疼了"],
"diarrhea": ["拉肚子", "腹泻", "拉稀"],
"gi-bleeding": ["黑便", "大便带血", "便血", "呕吐物带血"],
"constipation": ["便秘", "排便困难", "大便干结"],
"ibs": ["肚子胀气", "肠易激综合征", "腹痛腹胀"],
"ibd": ["溃疡性结肠炎", "克罗恩病", "肠道炎症"],
"stomach-cancer": ["胃不舒服", "胃炎", "胃病"],
"diabetes": ["血糖高", "血糖低", "头晕乏力", "手脚发麻", "低血糖"],
"osteoporosis": ["补钙", "骨头疼", "腰背痛", "骨质疏松", "摔伤", "骨折"],
}
TIME_CHRONIC = ["两周", "一个月", "三个月", "半年", "一年", "两个星期"]
REC = ["三", "四", "五六", "七八"]
DOMAIN_ENT = {
"gi-bleeding": ["大便发黑", "呕血", "便血", "柏油样便", "上消化道出血"],
"diabetes": ["出冷汗心慌手抖", "低血糖发作", "意识模糊", "酮症酸中毒", "血糖很高"],
"osteoporosis": ["摔了一跤髋部很疼", "跌倒后腰背剧痛", "骨折了", "剧烈骨痛"],
}
def gen_alert_realcat():
"""Red-flag utterances; label by rules but keep the explicit category by keyword."""
recs = []
# chronicity (grounded-entity that becomes alert by duration)
for cat, ents in CAT_SYN.items():
for e in ents:
if "孩子" in e:
continue
for t in ["我已经{e}{time}了,饮食上怎么办", "{e}一直反复{time}了", "{e}越来越严重,怎么调理"]:
time = random.choice(TIME_CHRONIC)
u = t.format(e=e, time=time)
plan = plan_of(u)
if plan["mode"] == "alert":
# force category to match the synthetic entity's domain
if not plan["sub_intents"]:
plan["sub_intents"] = [{"intent": "what-eat", "entity": "", "category": cat,
"level": "basics", "tmpl": "what-eat",
"risk": "high"}]
plan["sub_intents"][0]["category"] = cat
plan["sub_intents"][0]["intent"] = "what-eat" if "饮食" in u or "调理" in u else "diet-advice"
recs.append(build_example(u, plan))
# domain red-flag entities (force alert)
for cat, ents in DOMAIN_ENT.items():
for e in ents:
u = f"我{e},要紧吗,该怎么吃"
plan = plan_of(u)
if plan["mode"] == "alert":
if not plan["sub_intents"]:
plan["sub_intents"] = [{"intent": "diet-advice", "entity": "", "category": cat,
"level": "basics", "tmpl": "diet-advice",
"risk": "high"}]
plan["sub_intents"][0]["category"] = cat
recs.append(build_example(u, plan))
return recs
CAT_SYN = {
"gastritis": ["胃痛", "胃胀", "反酸", "胃炎"],
"peptic-ulcer": ["胃溃疡", "上腹痛"],
"diarrhea": ["拉肚子", "腹泻"],
"constipation": ["便秘", "排便困难"],
"stomach-cancer": ["胃不舒服"],
"diabetes": ["血糖高", "手脚发麻", "头晕乏力"],
"osteoporosis": ["骨头疼", "腰背痛"],
}
# ========================= 3) intent coverage =========================
ADVICE_TMPL = {
"what-eat": ["胃溃疡病人平时能吃些什么食物", "糖尿病人可以吃水果吗", "骨质疏松的人吃什么补钙食物好"],
"what-avoid": ["胃病有没有什么忌口", "糖尿病病人不能吃哪些食物", "骨质疏松要少吃什么"],
"diet-advice": ["腹泻之后饮食上怎么调理", "糖尿病肾病饮食注意什么", "老年人怎么通过饮食养护骨骼"],
"nutrient": ["补钙每天吃多少合适", "糖尿病人需要补充哪些维生素", "骨质疏松该补蛋白质还是钙"],
"meal-plan": ["糖尿病一日三餐怎么搭配", "骨质疏松的食谱怎么安排", "胃病病人一日三餐怎么吃"],
"lifestyle": ["糖尿病人平时该怎么运动", "补钙要配合晒太阳吗", "骨质疏松病人适合做什么运动"],
"definition": ["什么是肠易激综合征", "骨质疏松是怎么回事", "消化性溃疡是什么"],
"when-seek-care": ["胃痛什么情况要去看医生", "血糖多高需要去医院", "摔伤后要不要拍片"],
}
def gen_intent():
recs = []
for intent, us in ADVICE_TMPL.items():
for u in us:
plan = plan_of(u)
if plan["sub_intents"]:
plan["sub_intents"][0]["intent"] = intent
recs.append(build_example(u, plan))
return recs
# ========================= 4) compound =========================
COMPOUND = [
("胃痛和反酸,吃什么好", ["gastritis"], ["what-eat"]),
("糖尿病还便秘,饮食怎么办", ["diabetes", "constipation"], ["diet-advice"]),
("老人骨质疏松又胃不好,怎么补钙", ["osteoporosis", "gastritis"], ["nutrient"]),
("血糖高还经常拉肚子,怎么调理", ["diabetes", "diarrhea"], ["diet-advice"]),
("想了解骨质疏松,也想问怎么补钙", ["osteoporosis"], ["definition", "nutrient"]),
("胃溃疡何十二指肠溃疡都要注意什么", ["peptic-ulcer"], ["definition"]),
]
def gen_compound():
recs = []
for u, cats, ints in COMPOUND:
plan = plan_of(u)
if plan["sub_intents"]:
plan["sub_intents"] = [
{"intent": ints[0], "entity": "", "category": cats[0], "level": "basics",
"tmpl": ints[0], "risk": "high" if plan["mode"] == "alert" else "low"},
]
recs.append(build_example(u, plan))
return recs
def main():
random.seed(42)
recs = []
# 1) real curated
for key in ("gastro", "diabetes", "osteoporosis"):
for line in open(os.path.join(ROOT, "data", f"qa_{key}.jsonl"), encoding="utf-8"):
r = json.loads(line)
u = r["question"]
if not u or "。" not in u:
u = u.rstrip("。") + "怎么办"
plan = plan_of(u, override_cat=r["category"], override_intent=rule_intent(u))
recs.append(build_example(u, plan))
recs += gen_alert_realcat()
recs += gen_intent()
recs += gen_compound()
# dedup by utterance
seen = set()
uniq = []
for r in recs:
if r["utterance"] in seen:
continue
seen.add(r["utterance"])
uniq.append(r)
random.shuffle(uniq)
val_n = max(30, int(len(uniq) * 0.12))
val = uniq[:val_n]
train = uniq[val_n:]
with open(os.path.join(OUT, "router_sft_train.jsonl"), "w", encoding="utf-8") as f:
for r in train:
f.write(json.dumps(r, ensure_ascii=False) + "\n")
with open(os.path.join(OUT, "router_sft_val.jsonl"), "w", encoding="utf-8") as f:
for r in val:
f.write(json.dumps(r, ensure_ascii=False) + "\n")
from collections import Counter
modes = Counter(json.loads(r["messages"][1]["content"])["mode"] for r in uniq)
cats = Counter()
intents = Counter()
for r in uniq:
p = json.loads(r["messages"][1]["content"])
for s in p["sub_intents"]:
cats[s["category"]] += 1
intents[s["intent"]] += 1
print(f"total unique: {len(uniq)} train={len(train)} val={len(val)}")
print("modes:", dict(modes))
print("categories:", dict(cats.most_common()))
print("intents:", dict(intents.most_common()))
if __name__ == "__main__":
main()
|