Teerapol Saengsukhiran

Runing with transformer model loader on oobabooga/text-generation-webui problem

1

I cannot use taskweaver on api hosted by oobabooga/text-generation-webui. However, it work with llama cpp model loader on oobabooga/text-generation-webui which I cannot use it well as AMD rocm 5.7 has...

Rocm support

7

Please consider adding rocm support for amd gpu

please consider supporting amd gpu including consumer grade like gfx1100 (7900xtx)

please consider adding Rocm support for AMD RDNA3 + other amd gpu

Request for code in Appendix F

1

Can you provide the code you used for testing In appendix F? (Representation analysis) I want to test it with llama3 and mistral v 0.3

Please add rocm support

1

Please consider adding rocm support for amd gpu (consumer 7900xtx and MI series)

[Issue]: Ubuntu 24.04.1 LTS incompatibility with rocm 6.2

2

### Problem Description rocminfo show ROCk module is NOT loaded, possibly no GPU devices I previously used rocm 6.1 on Ubuntu 24.04.4 (previous version but the number is somehow larger...

Rocm(7900xtx) GPU fail

12


Question about deepspeed checkpoint loading

1

I tried to load Lora training adapters from Deepspeed checkpoint: dir: ``` ls Bunny/checkpoints-llama3-8b/bunny-lora-llama3-8b-attempt2/checkpoint-6000 total 696M -rw-r--r-- 1 [email protected] CU 775 Nov 18 11:03 adapter_config.json -rw-r--r-- 1 [email protected] CU 686M...

About continuous_training

3

With continuous_training on bunny VLM, do we still need to specify vision_tower path? If we do point to siglip path, will it use that untrained weight instead of vision_tower that...