Problemsby IbisFine-tuning OpenAI is costly and unpredictable
MLOps3 posts from 3 people3d active+22 posts in the last 7 days, 1 the 7 days before (steady)
Posts per day
The posts behind it
3, newest first[Bug]: LoRA + Gemma4 MTP speculative decoding crashes at startup — draft pre_projection shape mismatch...MTP assistant): backbone (target) `hidden_size = 2816`, draft `hidden_size = 1024`. The failing layer is the draft's `pre_projection`: and the draft forward: `hidden_states` is backbone-dim (2816), so the concat only reaches 5632 when `inputs_embeds` comes from the **target's...vllm-project/vllm · bbenshab1 repliesOpen ↗[Bug]: Whisper long-audio transcription loses words at the internal 30 s cut points in json/text (split_audio cuts inside speech, no overlap or merge)### Your current environment Environment (collect_env.py not captured - summary) - vLLM **v0.29.0** (image `vllm/vllm-openai:v0.29.0`), also checked on **v0.31.0**: byte-identical output on 794 recordings - Model: `KBLab/kb-whisper-large` @ `d5d5984b` (a Whisper-large fine-tune...vllm-project/vllm · bjolo1 reactions, 1 repliesOpen ↗...compared to Jev - I'm not saying they will survive, but the competition CURRENTLY sucks - and also they aren't actually cheaper if you need them at scale. OpenAI is coming in twice the cost and so is Cloudflare. Ad yes you can finetune, and yes I do, but it also SUCKS....Hacker News comments · neilellisOpen ↗
Companies and products named
About this problem
Evidence
3 posts from 3 people in 2 places, about 2 a week over 9 days. Mostly on vllm-project/vllm, Hacker News comments. Tools named alongside: vLLM, Cloudflare, NVIDIA, GitHub.
Frustration Frustration 2 of 3· Seen on GitHub issues, Hacker News
How it was grouped
Posts that state a pain and match the "fine-tuning" rule about OpenAI. First post Oct 10, 2026, latest Oct 18, 2026. Corroborated: 3 or more posts from 2 or more people or places. 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 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 NVIDIA is costly and unpredictableFine-tuning ChatGPT is costly and unpredictable
Other OpenAI problems
OpenAI users keep asking for helpRunning OpenAI locally hits memory and speed limitsOpenAI upgrades break working setupsTrouble with OpenAI chatTrouble with OpenAI responsesOpenAI fails to return valid structured outputTrouble with OpenAI astraTrouble with OpenAI customTrouble with OpenAI imageOpenAI tables suffer from small files and skewTrouble with OpenAI crashesTrouble with OpenAI calls
History
Fine-tuning OpenAI is costly and unpredictable (3 posts)