turboderp-org/exllamav2
1
0
Fork 0
mirror of https://github.com/turboderp-org/exllamav2.git synced 2026-04-20 14:29:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'refs/remotes/LlamaEnjoyer/add_more_args_to_humaneval' into dev

Browse Source
This commit is contained in:
turboderp
2024-09-29 12:52:52 +02:00
parent 5d4359317d b2af0bbad3
commit ef7cdda31c
1 changed files with 12 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
eval/humaneval.py
View File

@@ -1,5 +1,8 @@
from __future__ import annotations
import sys, os
import os
import sys
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from human_eval.data import write_jsonl, read_problems
from exllamav2 import model_init
@@ -21,7 +24,10 @@ parser.add_argument("--max_tokens", type = int, default = 768, help = "Max numbe
parser.add_argument("-pf", "--prompt_format", type = str, help = "Instruct format to apply. Default is raw completion (for base models) ")
parser.add_argument("-v", "--verbose", action = "store_true", help = "Spam completions to console while generating")
parser.add_argument("-e", "--eval", action = "store_true", help = "Run evaluation script on output file after sampling")
parser.add_argument("-temp", "--temperature", type = float, help = "Sampling temperature (0 for greedy), default: 0.6")
parser.add_argument("-temp", "--temperature", type = float, help = "Sampling temperature (0 for greedy), default: 0.6", default = 0.6)
parser.add_argument("--top_k", type = int, help = "Top-k sampling, default: 50", default = 50)
parser.add_argument("--top_p", type = float, help = "Top-p sampling, default: 0.6", default = 0.6)
parser.add_argument("-trp", "--token_repetition_penalty", type = float, help = "Token repetition penalty, default: 1.0", default = 1.0)
model_init.add_args(parser)
args = parser.parse_args()
@@ -118,10 +124,10 @@ generator = ExLlamaV2DynamicGenerator(
)
gen_settings = ExLlamaV2Sampler.Settings(
token_repetition_penalty = 1.0,
temperature = 0.6,
top_k = 50,
top_p = 0.6
token_repetition_penalty=args.token_repetition_penalty,
temperature=args.temperature,
top_k=args.top_k,
top_p=args.top_p
)
# Get problems