@mariozechner/pi 0.1.3 → 0.1.5

package/README.md

@@ -5,12 +5,12 @@ Quickly deploy LLMs on GPU pods from [Prime Intellect](https://www.primeintellec
 ## Installation
 
 ```bash
-npm install -g @badlogic/pi
+npm install -g @mariozechner/pi
 ```
 
 Or run directly with npx:
 ```bash
-npx @badlogic/pi
+npx @mariozechner/pi
 ```
 
 ## What This Is
@@ -314,4 +314,4 @@ Remember: Tool calling is still an evolving feature in the LLM ecosystem. What w
 - **Connection Refused**: Check pod is running and port is correct
 - **HF Token Issues**: Ensure HF_TOKEN is set before running setup
 - **Access Denied**: Some models (like Llama, Mistral) require completing an access request on HuggingFace first. Visit the model page and click "Request access"
-- **Tool Calling Errors**: See the Tool Calling section above - consider disabling it or using a different model
+- **Tool Calling Errors**: See the Tool Calling section above - consider disabling it or using a different model

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mariozechner/pi",
-  "version": "0.1.3",
+  "version": "0.1.5",
   "description": "CLI tool for managing vLLM deployments on GPU pods from Prime Intellect, Vast.ai, etc.",
   "main": "pi.js",
   "bin": {

package/pi.js

@@ -204,7 +204,7 @@ class PrimeIntellectCLI {
             console.error('');
             console.error('Options:');
             console.error('  --name <name>      Model alias (default: auto-generated)');
-            console.error('  --context <size>   Context window: 4k, 8k, 16k, 32k or 4096, 8192, etc (default: model default)');
+            console.error('  --context <size>   Context window: 4k, 8k, 16k, 32k, 64k, 128k or 4096, 8192, etc (default: model default)');
             console.error('  --memory <percent> GPU memory: 30%, 50%, 90% or 0.3, 0.5, 0.9 (default: 90%)');
             console.error('  --all-gpus         Use all GPUs with tensor parallelism (ignores --memory)');
             console.error('  --vllm-args        Pass remaining args directly to vLLM (ignores other options)');
@@ -719,15 +719,20 @@ class PrimeIntellectCLI {
         console.log('  pi prompt <name> <msg>             Chat with a model\n');
         console.log('Start Options:');
         console.log('  --name <name>      Model alias (default: auto-generated)');
-        console.log('  --context <size>   Context window: 4k, 16k, 32k (default: 8k)');
+        console.log('  --context <size>   Context window: 4k, 8k, 16k, 32k, 64k, 128k (default: model default)');
         console.log('  --memory <percent> GPU memory: 30%, 50%, 90% (default: 90%)');
-        console.log('  --all-gpus         Use all GPUs with tensor parallelism\n');
+        console.log('  --all-gpus         Use all GPUs with tensor parallelism');
+        console.log('  --vllm-args        Pass remaining args directly to vLLM\n');
         console.log('Utility:');
         console.log('  pi shell                           SSH into active pod');
 
-        console.log('\nQuick Example:');
+        console.log('\nQuick Examples:');
         console.log('  pi start Qwen/Qwen2.5-7B-Instruct --name qwen');
         console.log('  pi prompt qwen "What is 2+2?"');
+        console.log('\n  # Qwen3-Coder on 8xH200 with custom vLLM args:');
+        console.log('  pi start Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8 --name qwen-coder --vllm-args \\');
+        console.log('    --data-parallel-size 8 --enable-expert-parallel \\');
+        console.log('    --tool-call-parser qwen3_coder --enable-auto-tool-choice --max-model-len 200000');
 
         if (this.config.active && this.config.pods[this.config.active]) {
             console.log(`\nActive pod: ${this.config.active} (${this.config.pods[this.config.active].ssh})`);