Text Generation
Transformers
Chinese
English
character-level
chinese
gru
mamba
Mixture of Experts
titans
Instructions to use masterofthestars/tianxuan with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use masterofthestars/tianxuan with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="masterofthestars/tianxuan")# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("masterofthestars/tianxuan", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use masterofthestars/tianxuan with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "masterofthestars/tianxuan" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "masterofthestars/tianxuan", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/masterofthestars/tianxuan
- SGLang
How to use masterofthestars/tianxuan with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "masterofthestars/tianxuan" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "masterofthestars/tianxuan", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "masterofthestars/tianxuan" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "masterofthestars/tianxuan", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use masterofthestars/tianxuan with Docker Model Runner:
docker model run hf.co/masterofthestars/tianxuan
File size: 6,384 Bytes
db7bc06 | 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 | /**
* 梯度监控器
* 记录和分析梯度范数,检测梯度爆炸/消失
*/
import * as fs from "fs";
import * as path from "path";
export class GradientMonitor {
private gradNormHistory: number[] = [];
private gradNormMax: number = 0;
private gradNormMin: number = Infinity;
private totalNorms: number = 0;
private stepCounts: number = 0;
private warningThreshold: number = 5.0;
private dangerousThreshold: number = 10.0;
// 日志文件
private logFile: string;
constructor(
private warningCallback: (msg: string) => void = (msg) => {
console.warn(`[梯度预警] ${msg}`);
},
private dangerCallback: (msg: string) => void = (msg) => {
console.error(`[梯度爆炸] ${msg}`);
},
logFile: string = "gradient_logs.json"
) {
this.logFile = logFile;
}
/**
* 记录梯度范数
*/
record(norm: number): void {
if (!Number.isFinite(norm)) {
this.warningCallback("收到非有限梯度范数: NaN/Inf");
return;
}
this.gradNormHistory.push(norm);
this.gradNormMax = Math.max(this.gradNormMax, norm);
this.gradNormMin = Math.min(this.gradNormMin, norm);
this.totalNorms += norm;
this.stepCounts++;
// 检查梯度异常
if (norm > this.dangerousThreshold) {
this.dangerCallback(`[梯度爆炸警告] 步数${this.stepCounts}: 范数=${norm.toFixed(4)} > 阈值${this.dangerousThreshold}`);
this.saveLog(); // 记录异常
} else if (norm > this.warningThreshold) {
this.warningCallback(`[梯度预警] 步数${this.stepCounts}: 范数=${norm.toFixed(4)} > 阈值${this.warningThreshold}`);
this.saveLog(); // 记录预警
} else if (norm < 1e-6 && norm > 0) {
this.warningCallback(`[梯度消失预警] 步数${this.stepCounts}: 范数=${norm.toExponential(4)}`);
}
}
/**
* 获取统计数据
*/
getStatistics(): {
avg: number;
max: number;
min: number;
median: number;
variance: number;
sampleCount: number;
gradNormHistory: number[];
} {
const avg = this.stepCounts > 0 ? this.totalNorms / this.stepCounts : 0;
const sorted = [...this.gradNormHistory].sort((a, b) => a - b);
const mid = Math.floor(sorted.length / 2);
const median = sorted.length % 2 === 0 ? (sorted[mid - 1] + sorted[mid]) / 2 : sorted[mid];
const variance = this.calculateVariance(avg);
return {
avg,
max: this.gradNormMax,
min: this.gradNormMin === Infinity ? 0 : this.gradNormMin,
median,
variance,
sampleCount: this.stepCounts,
gradNormHistory: [...this.gradNormHistory],
};
}
/**
* 报告当前状态
*/
report(): void {
const stats = this.getStatistics();
if (stats.sampleCount === 0) {
console.log("\n=== 梯度监控报告(空数据)===\n");
return;
}
console.log("\n========== 梯度监控报告 ==========");
console.log(`训练步数: ${stats.sampleCount}`);
console.log(`平均梯度范数: ${stats.avg.toFixed(6)}`);
console.log(`最大梯度范数: ${stats.max.toFixed(6)}`);
console.log(`最小梯度范数: ${stats.min.toFixed(6)}`);
console.log(`中位梯度范数: ${stats.median.toFixed(6)}`);
console.log(`梯度范数方差: ${stats.variance.toFixed(6)}`);
// 评估标准
let status = "正常";
if (stats.max > this.dangerousThreshold) status = "危险 (严重梯度爆炸!)";
else if (stats.max > this.warningThreshold) status = "预警 (梯度膨胀)";
else if (stats.avg < 1e-4) status = "异常 (梯度消失!)";
else if (stats.avg < 0.01) status = "稳定" + (stats.variance > 0.1 ? " (高变异性)" : "");
console.log(`\n评估结果:${status}`);
if (stats.max > 1.0) {
const quality = stats.max > 10.0 ? "严重" : "需要";
console.log(`[系统] 梯度${quality}裁剪:最大梯度=${stats.max.toFixed(4)}`);
}
console.log(`\n建议: 理想梯度范数范围 0.1-1.0,方差 < 0.2`);
console.log("======================================\n");
this.saveLog();
}
/**
* 计算方差
*/
private calculateVariance(mean: number): number {
if (this.gradNormHistory.length < 2) return 0;
const sumSquaredDiff = this.gradNormHistory.reduce((sum, norm) => {
return sum + Math.pow(norm - mean, 2);
}, 0);
return sumSquaredDiff / this.gradNormHistory.length;
}
/**
* 保存日志
*/
saveLog(): void {
try {
const stats = this.getStatistics();
const data = {
timestamp: new Date().toISOString(),
gradientNorms: stats.gradNormHistory.slice(-1000), // 保存最近1000条
statistics: {
avg: stats.avg,
max: stats.max,
min: stats.min,
median: stats.median,
variance: stats.variance,
sampleCount: stats.sampleCount,
},
};
const filePath = path.resolve(this.logFile);
fs.writeFileSync(filePath, JSON.stringify(data, null, 2));
} catch (e) {
// 静默异常
}
}
/**
* 重置监控器
*/
reset(): void {
this.gradNormHistory = [];
this.gradNormMax = 0;
this.gradNormMin = Infinity;
this.totalNorms = 0;
this.stepCounts = 0;
this.saveLog();
}
/**
* 加载历史日志
*/
loadHistory(): void {
try {
if (fs.existsSync(this.logFile)) {
const data = JSON.parse(fs.readFileSync(this.logFile, "utf-8"));
if (data.gradientNorms && Array.isArray(data.gradientNorms)) {
this.gradNormHistory = [...data.gradientNorms];
this.recalculateStats();
console.log(`[梯度监控] 已加载历史日志 (${this.stepCounts} 个样本)`);
}
}
} catch (e) {
// 静默异常
}
}
/**
* 根据历史数据重新计算统计
*/
private recalculateStats(): void {
if (this.gradNormHistory.length === 0) return;
this.gradNormMax = Math.max(...this.gradNormHistory);
this.gradNormMin = Math.min(...this.gradNormHistory);
this.totalNorms = this.gradNormHistory.reduce((a, b) => a + b, 0);
this.stepCounts = this.gradNormHistory.length;
}
/**
* 获取实时梯度范数(最后一条)
*/
getLatestGradNorm(): number {
return this.gradNormHistory.length > 0
? this.gradNormHistory[this.gradNormHistory.length - 1]
: 0;
}
}
|