Issues

Agent-ready Qwen3.6-35B recipe recommends nvidia/Qwen3.6-35B-A3B-NVFP4, which silently breaks agentic tool-calling on GB10 (malformed tool-call syntax past a tool-count cliff)

#89

· chiptoe-svg opened

on Jul 21, 2026

Proposal: native Reachy Mini control and Conversation playbook for DGX Spark ARM64

#86

· am423 opened

on Jul 19, 2026

Unsloth playbook: test_unsloth.py trains a broken model (flat loss ~7) — unsloth/Phi-3.5-mini-instruct mirror has a defective config; swap to microsoft/Phi-3.5-mini-instruct

#85

· legolaz8451 opened

on Jul 17, 2026

Request: playbook for disaggregated / cross-node inference on DGX Spark

#82

· knachiketa04 opened

on May 12, 2026

Vibe Coding VS Code

#80

· CodysGuide2AI opened

on Apr 29, 2026

vLLM 26.02-py3: Llama-3.3-70B-Instruct-NVFP4 serves at ~4 tok/s on GB10 single Spark

#79

· srirsiva opened

on Apr 25, 2026

TRT-LLM 1.2.0rc6: Nemotron-3-Super-120B-A12B-NVFP4 hangs the GB10 kernel during weight mmap (single-Spark, playbook-endorsed model)

#78

· srirsiva opened

on Apr 25, 2026

vLLM 26.02-py3 rejects Nemotron-3-Super-120B-A12B-NVFP4: "ModelOpt currently only supports ['FP8', 'FP8_PER_CHANNEL_PER_TOKEN', 'FP8_PB_WO', 'NVFP4']" but checkpoint declares MIXED_PRECISION

#77

· srirsiva opened

on Apr 25, 2026

vLLM: Phi-4-reasoning-plus-NVFP4 emits garbage tokens on 26.02-py3 (chat template / tokenizer mismatch)

#76

· srirsiva opened

on Apr 25, 2026

Gemma4 cannot run with DGX Spark vLLM Ray playbook: NGC vLLM image supports Ray but does not recognize model_type gemma4; gemma4-cu130 image lacks ray

#75

· 2623674073 opened

on Apr 24, 2026

Ollama Tensor Parallelism

#72

· rumplestilzken opened

on Apr 14, 2026

Build and Deploy a Multi-Agent Chatbot - Build Failed (Frontend)

#71

· Billybiff opened

on Mar 29, 2026

Navigation Menu

Uh oh!

All issues

Agent-ready Qwen3.6-35B recipe recommends nvidia/Qwen3.6-35B-A3B-NVFP4, which silently breaks agentic tool-calling on GB10 (malformed tool-call syntax past a tool-count cliff)

Proposal: native Reachy Mini control and Conversation playbook for DGX Spark ARM64

Unsloth playbook: test_unsloth.py trains a broken model (flat loss ~7) — unsloth/Phi-3.5-mini-instruct mirror has a defective config; swap to microsoft/Phi-3.5-mini-instruct

Request: playbook for disaggregated / cross-node inference on DGX Spark

Vibe Coding VS Code

vLLM 26.02-py3: Llama-3.3-70B-Instruct-NVFP4 serves at ~4 tok/s on GB10 single Spark

TRT-LLM 1.2.0rc6: Nemotron-3-Super-120B-A12B-NVFP4 hangs the GB10 kernel during weight mmap (single-Spark, playbook-endorsed model)

vLLM 26.02-py3 rejects Nemotron-3-Super-120B-A12B-NVFP4: "ModelOpt currently only supports ['FP8', 'FP8_PER_CHANNEL_PER_TOKEN', 'FP8_PB_WO', 'NVFP4']" but checkpoint declares MIXED_PRECISION

vLLM: Phi-4-reasoning-plus-NVFP4 emits garbage tokens on 26.02-py3 (chat template / tokenizer mismatch)

Gemma4 cannot run with DGX Spark vLLM Ray playbook: NGC vLLM image supports Ray but does not recognize model_type gemma4; gemma4-cu130 image lacks ray

Ollama Tensor Parallelism

Build and Deploy a Multi-Agent Chatbot - Build Failed (Frontend)

Uh oh!

Search results