|
46 | 46 | "dataset": "wikitext2", |
47 | 47 | "awq": True, |
48 | 48 | }, |
49 | | - "meta-llama/Llama-2-7b": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.7}, |
| 49 | + "meta-llama/Llama-2-7b": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.6}, |
50 | 50 | "meta-llama/Llama-2-7b-chat": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.8}, |
51 | 51 | "meta-llama/Llama-2-13b-chat": {"bits": 4, "sym": True, "group_size": 64, "ratio": 0.8}, |
52 | | - "stabilityai/stablelm-3b-4e1t": {"bits": 4, "sym": True, "group_size": 64, "ratio": 0.8}, |
53 | | - "stablelm-epoch-3b-preview": {"bits": 4, "sym": True, "group_size": 64, "ratio": 0.8}, |
54 | | - "stabilityai/stablelm-zephyr-3b": {"bits": 4, "sym": False, "group_size": 128, "ratio": 1.0}, |
| 52 | + "stabilityai/stablelm-3b-4e1t": { |
| 53 | + "bits": 4, |
| 54 | + "sym": True, |
| 55 | + "group_size": 64, |
| 56 | + "ratio": 0.8, |
| 57 | + "dataset": "wikitext2", |
| 58 | + "awq": True, |
| 59 | + }, |
| 60 | + "stabilityai/stablelm-zephyr-3b": { |
| 61 | + "bits": 4, |
| 62 | + "sym": False, |
| 63 | + "group_size": 128, |
| 64 | + "ratio": 1.0, |
| 65 | + "dataset": "wikitext2", |
| 66 | + "awq": True, |
| 67 | + }, |
| 68 | + "stabilityai/stable-code-3b": {"bits": 4, "sym": True, "group_size": 64, "ratio": 0.8}, |
55 | 69 | "pansophic/rocket-3B": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.8}, |
56 | 70 | "THUDM/chatglm2-6b": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.72}, |
57 | 71 | "Qwen/Qwen-7B-Chat": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.6}, |
58 | 72 | "openlm-research/open_llama_3b": {"bits": 4, "sym": True, "group_size": 64, "all_layers": True}, |
59 | | - "tiiuae/falcon-7b": {"bits": 4, "sym": True, "group_size": 64, "all_layers": True}, |
60 | | - "psmathur/orca_mini_3b": {"bits": 4, "sym": True, "group_size": 64, "all_layers": True}, |
| 73 | + "tiiuae/falcon-7b-instruct": {"bits": 4, "sym": True, "group_size": 64, "all_layers": True}, |
| 74 | + "psmathur/orca_mini_3b": { |
| 75 | + "bits": 4, |
| 76 | + "sym": True, |
| 77 | + "group_size": 64, |
| 78 | + "all_layers": True, |
| 79 | + "dataset": "wikitext2", |
| 80 | + "awq": True, |
| 81 | + }, |
| 82 | + "bigscience/bloomz-560m": { |
| 83 | + "bits": 4, |
| 84 | + "sym": True, |
| 85 | + "group_size": 64, |
| 86 | + "ratio": 0.8, |
| 87 | + "dataset": "wikitext2", |
| 88 | + "awq": True, |
| 89 | + }, |
61 | 90 | "mistralai/Mixtral-8x7B-v0.1": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.8}, |
62 | 91 | "facebook/opt-2.7b": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.7}, |
63 | 92 | "togethercomputer/RedPajama-INCITE-Chat-3B-v1": {"bits": 4, "sym": False, "group_size": 128, "ratio": 0.8}, |
|
0 commit comments