File size: 2,860 Bytes
64ddf8d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 | import os
from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
LLAMA_MODELS = {
"meta-llama/Llama-2-7b-hf": {
"layers": 32,
"experiment": True,
},
"meta-llama/Llama-2-13b-hf": {
"layers": 40,
"experiment": True,
},
"meta-llama/Meta-Llama-3-8B": {
"layers": 32,
"llama2": False,
"experiment": True,
},
"meta-llama/Llama-3.1-8B": {
"layers": 32,
"llama2": False,
"experiment": False,
},
"meta-llama/Meta-Llama-3-70B": {
"layers": 80,
"llama2": False,
"base_dir": "/data/hugginface/hub",
},
"meta-llama/Llama-2-70b-hf": {
"layers": 80,
"base_dir": "/data/hugginface/hub",
},
"meta-llama/Meta-Llama-3-70B-Instruct": {
"layers": 80,
"base_dir": "/data/hugginface/hub",
"llama2": False,
},
"meta-llama/Meta-Llama-3.1-405B-Instruct": {
"layers": 126,
"base_dir": "/data/hugginface/hub",
"llama2": False,
},
"../models/Llama-2-13b-hf": {
"layers": 40,
"base_dir": "../models/Llama-2-13b-hf",
"llama2": False,
},
"../models/Qwen3-8B": {
"layers": 36,
"base_dir": "../models/Qwen3-8B",
"llama2": False,
},
"../models/Qwen3-4B": {
"layers": 36,
"base_dir": "../models/Qwen3-4B",
"llama2": False,
},
"../models/Llama-3.2-3B-Instruct": {
"layers": 28,
"base_dir": "../models/Llama-3.2-3B-Instruct",
"llama2": False,
},
"../models/Mistral-7B-Instruct-v0.3": {
"layers": 32,
"base_dir": "../models/Mistral-7B-Instruct-v0.3",
"llama2": False,
},
"../models/Llama-3.1-70B-Instruct": {
"layers": 80,
"base_dir": "../models/Llama-3.1-70B-Instruct",
"llama2": False,
},
}
VIT_OPENCLIP_MODELS = {
"laion/CLIP-ViT-B-32-laion2B-s34B-b79K": {"vlayers": 12, "tlayers": 12},
"laion/CLIP-ViT-H-14-laion2B-s32B-b79K": {"vlayers": 32, "tlayers": 24},
"laion/CLIP-ViT-L-14-laion2B-s32B-b82K": {"vlayers": 24, "tlayers": 12},
}
SENSITIVITY_MODELS = [
"meta-llama/Llama-2-7b-hf",
"meta-llama/Llama-2-7b-chat-hf",
"epfl-llm/meditron-7b",
"meta-llama/Llama-3.1-8B",
"meta-llama/Llama-3.1-8B-Instruct",
]
def get_hf_model_storge_base_dir(model_id, hf_hub_dir=None):
# model_id_x = model_id.replace("/", "--")
# cache_dir = hf_hub_dir if hf_hub_dir else HUGGINGFACE_HUB_CACHE
# hf_model_dir = os.path.join(cache_dir, f"models--{model_id_x}")
# ref_main_fp = os.path.join(hf_model_dir, "refs", "main")
# with open(ref_main_fp) as fh:
# commit_sha = fh.read().strip()
# return os.path.join(hf_model_dir, "snapshots", commit_sha)
return os.path.join('../models', model_id.split("/")[-1])
|