some suggestions for actual use
I tested it for coding, and it works well and does a good job.but it. but this model tends to rewrite the whole file instead of just editing it. The rewriting often causes new issues, hope it can be optimized in the future. thank for your good job.
I've been using it with OMP successfully. It can surgically edit source files without any issues (using OMP's hashline).
I've been using it with OMP successfully. It can surgically edit source files without any issues (using OMP's hashline).
Interesting, ive been having massive issues with it inside of OMP. (v17.0.5). It fails to call tools, cant read artefacts, it tried 8 times to use read - only to fall back to cat + grep for a sizeable .md file, then that truncated and it couldnt expand/read further.
In Pi it does OK, but it still tends to hit a lot of snags.
Out of curiosity, how are you serving it and with what settings?
I've been using it with OMP successfully. It can surgically edit source files without any issues (using OMP's hashline).
Interesting, ive been having massive issues with it inside of OMP. (v17.0.5). It fails to call tools, cant read artefacts, it tried 8 times to use read - only to fall back to cat + grep for a sizeable .md file, then that truncated and it couldnt expand/read further.
In Pi it does OK, but it still tends to hit a lot of snags.
Out of curiosity, how are you serving it and with what settings?
llama.cpp/build/bin/llama-server \
-hf poolside/Laguna-XS-2.1-GGUF:Q4_K_M \
--host 0.0.0.0 \
--parallel 1 \
--batch-size 1024 \
--ubatch-size 512 \
--temp 1.0 \
--top-p 1.0 \
--top-k 20 \
--metrics \
-lv 4 \
--flash-attn on \
--jinja \
--cont-batching
But I've been using it with OMP 16, I am not sure if it might be an issue with the new xd:// protocol that was introduced in OMP 17.
But I've been using it with OMP 16, I am not sure if it might be an issue with the new xd:// protocol that was introduced in OMP 17.
Thanks for sharing, i also found your comment from another model discussion shortly after commenting here hehe.
And you may be on to something there, the new "xd ://" situation may be what is causing a lot of troubles, but even in my own benchmarks the Laguna XS 2.1 Q8 hit 17% on terminal-bench2 (which is ok) and i am currently still running it through some isolated benchmarks with simulated captured issues and bugs in my own environment - but it tends to loop a lot, even with the changed temps (tried 0.0, 0.2, 0.6, 0.8 and 1.0).
It does a pretty good job at creating frontend stuff, like dashboards or GUIs - which i didnt expect.