Browse categories
Explore
Fiverr Pro
English
$
USD
I will fix python bugs in pytorch and hugging face transformers
Junling G
About this gig
Stuck on a Python traceback, a PyTorch tensor error, or a Hugging Face model that will not load? I will reproduce the agreed issue, fix the code or configuration, and explain how to verify the result.
I am an AI engineering student and open-source contributor. My contribution to Hugging Face Transformers, PR #45402, fixed registered-buffer loading under DeepSpeed ZeRO-3 and added a regression test. It was merged after maintainer review.
I can help with:
- Python exceptions and dependency conflicts
- PyTorch shape, dtype and device mismatches
- Transformers model, tokenizer and checkpoint loading
- Reproducible inference or training-script errors
You receive a scoped fix, corrected files or patch, and verification notes. Premium adds a regression test and walkthrough.
Please message me before ordering with the full traceback, package versions, hardware, and a minimal code sample. I will confirm feasibility and the right package.
One order covers one agreed issue. New features, full applications, large training runs and hardware upgrades are excluded. Cloud/GPU fees are not included. Distributed or hardware-specific issues require a custom scope.
Get to know Junling G
Junling G
AI Developer and Python Specialist
- FromChina
- Member sinceApr 2025
- Avg. response time1 hour
Languages
Chinese, English
Hi, I’m Junling, an AI engineering student specializing in local LLM deployment, fine-tuning, RAG, and AI agents.
I work with Qwen, Llama, Mistral, Ollama, Hugging Face, and PyTorch. I can help you:
* Install and optimize local AI models
* Create Python automation tools
I use a local RTX 40-series workstation, so your data can remain private and you can avoid unnecessary cloud costs.
Send me your requirements, hardware specifications, and expected outcome, and I’ll recommend the most practical solution.
My Portfolio
FAQ
What should I send before ordering?
Send the full error traceback, expected result, Python/library versions, OS, GPU/VRAM if relevant, and the smallest code sample that reproduces the issue. Remove passwords, API keys and private data. I will confirm feasibility and package scope first.
What does the Basic package cover?
One agreed Python error in one small script, up to 150 lines of relevant code. Includes the correction and a short explanation. Model pipelines, distributed training, new features and repository-wide debugging need a larger or custom scope. Please message me first.
Do you need access to my GPU or private data?
I start with anonymized code and sample data. Hardware-specific issues may require a test in your environment, agreed before ordering. Cloud/GPU charges are not included. Do not send passwords or API keys. I will not publish your code or data without permission.
What counts as a revision and a completed fix?
One revision covers corrections to the agreed issue in the agreed environment. We define a reproducible input and expected result before work starts. A new error, feature, model or environment is a separate scope. I do not promise that every issue is fixable; feasibility is checked first.
