# I need to fine tune Qwen3.5-9B with QLora on about 500 GB of raw data (images/scanned documents/pdfs/word/excel) - what is the best way forward? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/f360ecbd6b34ec85

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-25 22:32 ingested 2026-10-26 02:40

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xffuvl/i_need_to_fine_tune_qwen359b_with_qlora_on_about/)

## retained text



```
Hi All:

I done overthought and burned money after consulting with AI chats (from chatgpt to grok to deepseek and kimi/etc). I had to learn. 

I have in my possession:

\- 1 Asus ESC4000 G3 with 128 GB DDR4 RDIMM and in it 8x T4s deployment. This server hosts the Qwen 3.5-9B Q5\_K\_XL at 87k context per slot and 44 slots. Each T4 has its own llama.cpp instance and the average single user gets 35-40 tps. TTFT is about 3-5 seconds depending on the size of the context. Using a round robin nginx load balancer for requests. Extensive real time testing inside my own harness (built strictly for us, non coding use), it actually isn’t bad at all. (Comparing to gpt-4o-mini quality).

\- 2x V620s and 2 Mi50s (32GBs) and 1 3060 12 GB with another 3060 12 GB on the way (marketplace $250 cant complain tbh).

All this ran me $650 for Asus server, 350 per V620 and 480 shipped per Mi50

And now we get to the dumb part.

I got a ZP9E-D8 WS with 256GB DDR3 RDIMM for $300. 4 x16 slots - but we’re in the middle of a fight because nothing boots with even a single 32GB GPU. Fixing this i made a thread.

My wet dream of 2x Mi50 and 2x V620 node dried up real quick.

I also possess a T5610 with 64 GB DDR3 RDIMM, annnd only 1 Mi50/V620 can be used because second GPU just wont boot. (Front panel code 2).

Then I have an B660M with 32 GB RAM and this puppy is able to post 2x V620 (hurray).

And finally, a ROG X299 E-Gaming with 3 PCIE slots 2 at x16, 1 x8 and 1 x4 - i can only boot 1 Mi50 in it :/

So, with all that said. I am beyond lost on what to do, I need to fine tune my model and don’t know what path to take. Please help, I am beyond overwhelmed. I don’t wanna waste anymore money. I just want to fine tune. I’m thinking sticking with 3060 12 GBs but then the appeal of a single R9700 is so great that I am sick in my stomach. I don’t know at all if I can fine tune with unsloth + intel GPUs otherwise I’d go B65/B70 in a heartbeat (i think they’re like $700).

Please help people.
```



document fetched 2026-10-26 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xffuvl/i_need_to_fine_tune_qwen359b_with_qlora_on_about
