| Welcome, Guest |
You have to register before you can post on our site.
|
| Online Users |
There are currently 191 online users.
» 0 Member(s) | 188 Guest(s)
Applebot, Bing, Google
|
| Latest Threads |
FC 27 Season 2 Coins Guid...
Forum: Off Topic
Last Post: salisy
Yesterday, 01:50 AM
» Replies: 0
» Views: 20
|
Aion 2 October 21 Update ...
Forum: Off Topic
Last Post: salisy
Yesterday, 01:33 AM
» Replies: 0
» Views: 26
|
CoD: Modern Warfare 4 Is ...
Forum: Off Topic
Last Post: salisy
Yesterday, 01:18 AM
» Replies: 0
» Views: 22
|
consumer accelerators wor...
Forum: CPUs / NPUs / Accelerators
Last Post: rocmcrate26
10-19-2026, 05:59 AM
» Replies: 11
» Views: 291
|
structured output failure...
Forum: Structured Output
Last Post: latentpeak22
10-19-2026, 05:59 AM
» Replies: 13
» Views: 410
|
CUDA update broke my thro...
Forum: NVIDIA / CUDA
Last Post: nanobrook
10-19-2026, 05:58 AM
» Replies: 13
» Views: 377
|
70B still worth the pain
Forum: Proprietary / Frontier
Last Post: routerunner
10-19-2026, 05:58 AM
» Replies: 15
» Views: 408
|
context size vs speed in ...
Forum: llama.cpp / GGUF
Last Post: macroforge
10-19-2026, 05:58 AM
» Replies: 15
» Views: 428
|
do small edge boxes make ...
Forum: CPUs / NPUs / Accelerators
Last Post: kernelmike
10-19-2026, 05:58 AM
» Replies: 15
» Views: 436
|
8B vs 14B sweet spot: any...
Forum: Proprietary / Frontier
Last Post: forgemixer
10-19-2026, 05:57 AM
» Replies: 8
» Views: 263
|
|
|
| benchmarks that map to real use |
|
Posted by: routeline23 - 10-08-2026, 08:46 PM - Forum: Evals
- Replies (10)
|
 |
Thought I'd post a small recipe because a lot of threads ask similar things.
Maybe useful as a starting point: Ollama on 7900 XTX, keeping reranking in mind from the start.
Maybe useful as a starting point for newer folks.
I care more about stability than squeezing the last possible tokens/sec out of it.
|
|
|
| setup notes worth sharing: early impressions |
|
Posted by: mintstation99 - 10-08-2026, 08:46 PM - Forum: Local
- Replies (7)
|
 |
Did a bit of testing and thought I'd post the result here.
I tried this with Phi using vLLM on mini PC. The headline result was better than I expected, but MoE serving still ended up being the limiting factor.
Curious if anyone else landed in the same place.
|
|
|
|
| prefill bottlenecks on longer prompts on A6000 |
|
Posted by: batchscope72 - 10-08-2026, 08:26 PM - Forum: Other Engines
- Replies (10)
|
 |
Thought I'd post a small recipe because a lot of threads ask similar things.
Just sharing a setup that ended up working reasonably well: Gemma, MLX, 4090, with a lot of attention on MoE serving.
If someone wants the exact settings I can clean them up and post them.
|
|
|
|
| system prompt habits that helped for Mixtral |
|
Posted by: mintstation99 - 10-08-2026, 07:42 PM - Forum: Prompts / System Prompts
- Replies (13)
|
|
Maybe I'm missing something obvious.
I'm trying to decide whether to keep the current stack or change course. Main question is how people here are handling VRAM headroom.
If you've tested something similar, what did you keep and what did you stop bothering with?
This is mostly for local work rather than public serving.
|
|
|
|
| ROCm version that finally behaves |
|
Posted by: thetaember38 - 10-08-2026, 07:28 PM - Forum: AMD / ROCm
- Replies (11)
|
 |
Did a bit of testing and thought I'd post the result here.
Short version: TensorRT-LLM behaved well for me, but only after I stopped ignoring tokens/sec.
Happy to post more details if anyone wants them.
|
|
|
|
| CLI helpers for local models with MLX |
|
Posted by: batchshed - 10-08-2026, 06:39 PM - Forum: Local
- Replies (12)
|
 |
This keeps coming up in conversation so I figured I'd start a thread.
The more I work with local setups, the more it seems like VRAM headroom decides everything downstream.
Curious whether that lines up with what other people are seeing.
|
|
|
|
| CLI helpers for local models for Gemma |
|
Posted by: modgarden - 10-08-2026, 06:32 PM - Forum: Local
- Replies (12)
|
 |
Has anyone else run into this?
I'm mostly working with DeepSeek in llama.cpp on M3 Max. The part I'm trying to understand is NVFP4.
Not looking for the perfect answer, mostly what worked in real use.
This is mostly for local work rather than public serving.
|
|
|
|
| Tensor cores actually helping here |
|
Posted by: mintforge - 10-08-2026, 06:00 PM - Forum: NVIDIA / CUDA
- Replies (1)
|
 |
Spent some time on this today and wanted to compare notes.
Two paths both seem viable, but they break down differently once HBM constraints becomes important.
What would you pick if you had to live with it for a few months?
|
|
|
|
| synthetic data for support bots: anyone testing this? |
|
Posted by: routeline23 - 10-08-2026, 05:54 PM - Forum: Routers
- Replies (11)
|
|
Thought I'd post a small recipe because a lot of threads ask similar things.
Just sharing a setup that ended up working reasonably well: Llama, llama.cpp, B200, with a lot of attention on tokens/sec.
If someone wants the exact settings I can clean them up and post them.
|
|
|
|
| flux pipeline local tips with llama.cpp |
|
Posted by: cobaltspark - 10-08-2026, 05:36 PM - Forum: Model Comparisons
- Replies (1)
|
 |
Did a bit of testing and thought I'd post the result here.
After a few runs on mini PC, I ended up with decent performance, but the setup only really made sense once I adjusted around structured output.
Happy to post more details if anyone wants them.
This is mostly for local work rather than public serving.
If I had to summarize the lesson so far, it would be that clean setup and observability beat clever guesses.
|
|
|
|
|
