forked from avnermay/Sequoia
-
Notifications
You must be signed in to change notification settings - Fork 0
/
tree_search.sh
90 lines (79 loc) · 5.66 KB
/
tree_search.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
# 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
export ACC_RATE_DIR=/work/avner/results/sequoia/acceptance_rates
export SPEED_DIR=/work/avner/results/sequoia/forward_pass_times
export OUTPUT_DIR=/work/avner/results/sequoia/growmaps
# -ntasks=1 --cpus-per-task=4 --mem=20G
# models: Llama-2-7b/70b-chat 16 bit
# dataset: oasst
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/tree_search.py \
--output_path $OUTPUT_DIR/oasst_target_meta-llama--Llama-2-70b-chat-hf_draft_meta-llama--Llama-2-7b-chat-hf_temp_0.6_top_p_0.9_growmap \
--acceptance_rates $ACC_RATE_DIR/oasst_target_meta-llama--Llama-2-70b-chat-hf_draft_meta-llama--Llama-2-7b-chat-hf_temp_0.6_top_p_0.9_stochastic_acc_rates_fast.json \
--target_model_speeds $SPEED_DIR/meta-llama--Llama-2-70b-chat-hf_forward_pass_times.json \
--draft_model_speeds $SPEED_DIR/meta-llama--Llama-2-7b-chat-hf_forward_pass_times.json"
sbatch \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/tree_search.py \
--output_path $OUTPUT_DIR/oasst_target_meta-llama--Llama-2-70b-chat-hf_draft_meta-llama--Llama-2-7b-chat-hf_temp_0.05_top_p_1.0_growmap \
--acceptance_rates $ACC_RATE_DIR/oasst_target_meta-llama--Llama-2-70b-chat-hf_draft_meta-llama--Llama-2-7b-chat-hf_temp_0.05_top_p_1.0_stochastic_acc_rates_fast.json \
--target_model_speeds $SPEED_DIR/meta-llama--Llama-2-70b-chat-hf_forward_pass_times.json \
--draft_model_speeds $SPEED_DIR/meta-llama--Llama-2-7b-chat-hf_forward_pass_times.json"
sbatch \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
# models: Llama-2-7b/70b 16bit NON-CHAT!
# dataset: Wikitext
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/tree_search.py \
--output_path $OUTPUT_DIR/wikitext_target_meta-llama--Llama-2-70b-hf_draft_meta-llama--Llama-2-7b-hf_temp_0.6_top_p_0.9_growmap \
--acceptance_rates $ACC_RATE_DIR/wikitext_target_meta-llama--Llama-2-70b-hf_draft_meta-llama--Llama-2-7b-hf_temp_0.6_top_p_0.9_stochastic_acc_rates_fast.json \
--target_model_speeds $SPEED_DIR/meta-llama--Llama-2-70b-hf_forward_pass_times.json \
--draft_model_speeds $SPEED_DIR/meta-llama--Llama-2-7b-hf_forward_pass_times.json"
sbatch \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/tree_search.py \
--output_path $OUTPUT_DIR/wikitext_target_meta-llama--Llama-2-70b-hf_draft_meta-llama--Llama-2-7b-hf_temp_0.05_top_p_1.0_growmap \
--acceptance_rates $ACC_RATE_DIR/wikitext_target_meta-llama--Llama-2-70b-hf_draft_meta-llama--Llama-2-7b-hf_temp_0.05_top_p_1.0_stochastic_acc_rates_fast.json \
--target_model_speeds $SPEED_DIR/meta-llama--Llama-2-70b-hf_forward_pass_times.json \
--draft_model_speeds $SPEED_DIR/meta-llama--Llama-2-7b-hf_forward_pass_times.json"
sbatch \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
# models: Vicuna-33B + SL1.3B
# dataset: oasst
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/tree_search.py \
--output_path $OUTPUT_DIR/oasst_target_lmsys--vicuna-33b-v1.3_draft_princeton-nlp--Sheared-LLaMA-1.3B_temp_0.6_top_p_0.9_growmap \
--acceptance_rates $ACC_RATE_DIR/oasst_target_lmsys--vicuna-33b-v1.3_draft_princeton-nlp--Sheared-LLaMA-1.3B_temp_0.6_top_p_0.9_stochastic_acc_rates_fast.json \
--target_model_speeds $SPEED_DIR/lmsys--vicuna-33b-v1.3_forward_pass_times.json \
--draft_model_speeds $SPEED_DIR/princeton-nlp--Sheared-LLaMA-1.3B_forward_pass_times.json"
sbatch \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm
CMD="/home/avner/anaconda3/envs/env2/bin/python /work/avner/git2/Sequoia/tree_search.py \
--output_path $OUTPUT_DIR/oasst_target_lmsys--vicuna-33b-v1.3_draft_princeton-nlp--Sheared-LLaMA-1.3B_temp_0.05_top_p_1.0_growmap \
--acceptance_rates $ACC_RATE_DIR/oasst_target_lmsys--vicuna-33b-v1.3_draft_princeton-nlp--Sheared-LLaMA-1.3B_temp_0.05_top_p_1.0_stochastic_acc_rates_fast.json \
--target_model_speeds $SPEED_DIR/lmsys--vicuna-33b-v1.3_forward_pass_times.json \
--draft_model_speeds $SPEED_DIR/princeton-nlp--Sheared-LLaMA-1.3B_forward_pass_times.json"
sbatch \
--export=HF_HOME,HF_DATASETS_CACHE,TRANSFORMERS_CACHE,DATA_DIR,WANDB_CACHE_DIR,WANDB_DATA_DIR,WANDB_ARTIFACT_DIR,CMD_TO_RUN="$CMD" \
/work/avner/git/spec-dec/launch_scripts/launch_job.slurm