forked from Infini-AI-Lab/Sequoia
-
Notifications
You must be signed in to change notification settings - Fork 1
/
measure_acceptance_rates.sh
113 lines (103 loc) · 4.23 KB
/
measure_acceptance_rates.sh
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
102
103
104
105
106
107
108
109
110
111
112
113
# models: Llama-2-7b/70b-chat 16 bit
# dataset: oasst
# - /work/avner/git/speculative/data/oasst_prompts.json
# - [0:100] - test; [100: 540] - validation
# - length range 140 - 1206 tokens (test)
# temperatures:
# - t 0.6, top_p 0.9
# - t 0.0
# models: Llama-2-7b/70b 16bit NON-CHAT!
# dataset: Wikitext [considered C4 but prefer to stick with WT]
# - /work/avner/git/speculative/data/wikitext_prompts.json
# temperatures:
# - t 0.6, top_p 0.9
# - t 0.0
# models: Vicuna-33B + SL1.3B
# dataset: oasst, length range (200 - 1200+)
# - /work/avner/git/speculative/data/oasst_prompts.json
# - [0:100] - test; [100: 540] - validation
# temperatures:
# - t 0.6, top_p 0.9
# - t 0.0
# [Optional: ]
# models: Llama-2-7b/70b-chat 16 bit
# dataset: mtbench (80 entries)
# - sequoia/tests/dataset/mt_bench.jsonl"
# temperatures:
# - t 0.6, top_p 0.9
# - t 0.0
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/measure_acceptance_rates.py \
--draft meta-llama/Llama-2-7b-chat-hf \
--target meta-llama/Llama-2-70b-chat-hf \
--dataset oasst \
--temp 0.6 \
--top_p 0.9 \
--algorithm stochastic \
--end 412 \
--offloading \
--output_dir /work/avner/results/sequoia"
sbatch --gpus=8 --nodelist=mk-viii-06 \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/measure_acceptance_rates.py \
--draft meta-llama/Llama-2-7b-chat-hf \
--target meta-llama/Llama-2-70b-chat-hf \
--dataset oasst \
--algorithm greedy \
--temp 0.05 \
--end 412 \
--offloading \
--output_dir /work/avner/results/sequoia"
sbatch --gpus=8 \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/measure_acceptance_rates.py \
--draft meta-llama/Llama-2-7b-hf \
--target meta-llama/Llama-2-70b-hf \
--dataset wikitext \
--temp 0.6 \
--top_p 0.9 \
--algorithm stochastic \
--end 412 \
--offloading \
--output_dir /work/avner/results/sequoia"
sbatch --gpus=8 --nodelist=mk-viii-08 \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/measure_acceptance_rates.py \
--draft meta-llama/Llama-2-7b-hf \
--target meta-llama/Llama-2-70b-hf \
--dataset wikitext \
--algorithm greedy \
--temp 0.05 \
--end 412 \
--offloading \
--output_dir /work/avner/results/sequoia"
sbatch --gpus=8 \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/measure_acceptance_rates.py \
--draft princeton-nlp/Sheared-LLaMA-1.3B \
--target lmsys/vicuna-33b-v1.3 \
--dataset oasst \
--temp 0.6 \
--top_p 0.9 \
--algorithm stochastic \
--end 412 \
--offloading \
--output_dir /work/avner/results/sequoia"
sbatch --gpus=8 --nodelist=mk-viii-09 \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/measure_acceptance_rates.py \
--draft princeton-nlp/Sheared-LLaMA-1.3B \
--target lmsys/vicuna-33b-v1.3 \
--dataset oasst \
--algorithm greedy \
--temp 0.05 \
--end 412 \
--offloading \
--output_dir /work/avner/results/sequoia"
sbatch --gpus=8 \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm