Problemsby IbisFine-tuning NVIDIA is costly and unpredictable
MLOps2 posts from 2 people2d active+00 posts in the last 7 days, 0 the 7 days before (fading)Unverified
Posts per day
The posts behind it
2, newest firstParakeet TDT greedy decoding ignores `max_symbols_per_step`: unbounded token repeats on one encoder frame### System Info - transformers 5.15.1 (the relevant code is unchanged on `main` as of 5.18.0) - torch 2.8.0, Python 3.13; macOS (CPU) and Linux (CUDA) behave the same - model: `nvidia/parakeet-tdt-0.6b-v3` ### Who can help? @eustlb (Parakeet / `generation_parakeet.py`) ###...huggingface/transformers · tomschott2 repliesOpen ↗[Bug]: Fill-in DSpark underallocates KV lookahead and can overwrite target KV (synthetic CUDA repro)...× 20 batches × 16 requests completed without the observed corruption. The originally failing batch had no invalid draft writes or target-KV mutations. ### Proposed fixes - #59105: minimal two-file fix and regression test; reserve N+1 slots for fill-in DSpark while preserving...vllm-project/vllm · notimesea1 reactionsOpen ↗
Companies and products named
About this problem
Evidence
2 posts from 2 people in 2 places, about 1 a week over 10 days. Mostly on vllm-project/vllm, huggingface/transformers. Tools named alongside: Codex, vLLM, PyTorch, GitHub.
Frustration Frustration 0 of 3· Seen on GitHub issues
How it was grouped
Posts that state a pain and match the "fine-tuning" rule about NVIDIA. First post Sep 28, 2026, latest Oct 7, 2026. Unverified: fewer than 3 posts, or one person or place only. Method
The same problem elsewhere
Fine-tuning Hugging Face is costly and unpredictableFine-tuning is costly and results are unpredictableFine-tuning Qwen is costly and unpredictableFine-tuning llama.cpp is costly and unpredictableFine-tuning OpenAI is costly and unpredictableFine-tuning vLLM is costly and unpredictableFine-tuning GitHub is costly and unpredictableFine-tuning Claude Code is costly and unpredictableFine-tuning Gemini is costly and unpredictableFine-tuning PyTorch is costly and unpredictableFine-tuning Claude is costly and unpredictableFine-tuning ChatGPT is costly and unpredictable
Other NVIDIA problems
NVIDIA GPU access and drivers are painfulNVIDIA runs out of memoryNVIDIA users keep filing bugsNVIDIA vulnerabilities are hard to triage and patchRunning NVIDIA locally hits memory and speed limitsNVIDIA upgrades break working setupsTrouble with NVIDIA accountNVIDIA lock-in makes switching costlyTrouble with NVIDIA inductorTrouble with NVIDIA phoneNVIDIA error messages are crypticTrouble with NVIDIA careers
History
steady to fading
Fine-tuning NVIDIA is costly and unpredictable (2 posts)