mirror of
https://github.com/turboderp-org/exllamav3.git
synced 2026-05-05 21:52:24 +00:00
20 lines
606 B
JSON
20 lines
606 B
JSON
[
|
|
{
|
|
"model_dir": "/mnt/str/models/llama3.1-8b-instruct/aqlm/2bit-1x16-g8",
|
|
"load_fn": "transformers",
|
|
"fwd_fn": "transformers",
|
|
"label": "AQLM 2bit 1x16-g8"
|
|
},
|
|
{
|
|
"model_dir": "/mnt/str/models/llama3.1-8b-instruct/aqlm/2bit-1x16-g8-pv",
|
|
"load_fn": "transformers",
|
|
"fwd_fn": "transformers",
|
|
"label": "AQLM 2bit 1x16-g8-pv"
|
|
},
|
|
{
|
|
"model_dir": "/mnt/str/models/llama3.1-8b-instruct/aqlm/2bit-2x8-g8-pv",
|
|
"load_fn": "transformers",
|
|
"fwd_fn": "transformers",
|
|
"label": "AQLM 2bit 2x8-g8-pv"
|
|
}
|
|
] |