#!/bin/bash BEAKER_IMAGE=${1:-nathanl/open_instruct_auto} # data mixes nonreasoner_integration_mix_decon="hamishivi/rlvr_acecoder_filtered_filtered 20000 hamishivi/omega-combined-no-boxed_filtered 20000 hamishivi/rlvr_orz_math_57k_collected_filtered 14000 hamishivi/polaris_53k 14000 hamishivi/MathSub-30K_filtered 9000 hamishivi/DAPO-Math-17k-Processed_filtered 7000 allenai/IF_multi_constraints_upto5_filtered_dpo_0625_filter-keyword-filtered-topic-char-topic-filtered 38000 allenai/rlvr_general_mix-keyword-filtered-topic-chars-char-filt-topic-filtered 50000" #hamishivi/tulu_3_rewritten_400k_string_f1_only_v2_nocode_all_filtered_qwen2_5_openthoughts2_filtered 22000 hamishivi/virtuoussy_multi_subject_rlvr_filtered 20000 hamishivi/new-wildchat-english-general_filtered 19000" # eval suite general_evals_int="gpqa:0shot_cot::qwen3-instruct,codex_humanevalplus:0-shot-chat::tulu-thinker_deepseek,alpaca_eval_v3::hamish_zs_reasoning_deepseek,ifeval::hamish_zs_reasoning_deepseek,agi_eval_english:0shot_cot::hamish_zs_reasoning_deepseek,omega_500:0-shot-chat_deepseek,minerva_math_500::hamish_zs_reasoning_deepseek,livecodebench_codegeneration::tulu-thinker_deepseek_no_think_tags_lite,aime:zs_cot_r1::pass_at_32_2024_deepseek,aime:zs_cot_r1::pass_at_32_2025_deepseek,zebralogic::hamish_zs_reasoning_deepseek,bbh:cot::hamish_zs_reasoning_deepseek_v2,mmlu:cot::hamish_zs_reasoning_deepseek,popqa::hamish_zs_reasoning_deepseek,mbppplus:0-shot-chat::tulu-thinker_deepseek" # model checkpoint model_name_or_path="/weka/oe-adapt-default/scottg/olmo/merging/ckpts/olmo3-instruct-dpo-1116-vibes/olmo3-7b-DPO-1115-newb-tpc-d5-lbc100-bal-1e-6-1__42__1763293644" # nov 16 tentative final checkpoint row 33 # was replace by row 27 # cluster cluster=ai2/jupiter #template chat_template=tokenizer_default NUM_GPUS=${NUM_GPUS:-8} hosted_vllm="" exp_name="grpo_math_only_p64_4_8k_${gs_model_name}" gs_model_name="olmo3-instruct-dpo-hpz1" EXP_NAME=${EXP_NAME:-${exp_name}} uv run python mason.py \ --description $exp_name \ --task_name ${EXP_NAME} \ --cluster ${cluster} \ --workspace ai2/olmo-instruct \ --priority urgent \ --pure_docker_mode \ --image $BEAKER_IMAGE \ --preemptible \ --num_nodes 8 \ --max_retries 5 \ --env VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \ --env HOSTED_VLLM_API_BASE=$hosted_vllm \ --gs_model_name $gs_model_name \ --gpus ${NUM_GPUS} \ --budget ai2/oe-other -- source configs/beaker_configs/ray_node_setup.sh \&\& source configs/beaker_configs/code_api_setup.sh \&\& python open_instruct/grpo_fast.py \ --exp_name ${EXP_NAME} \ --beta 0.0 \ --num_samples_per_prompt_rollout 8 \ --num_unique_prompts_rollout 64 \ --num_mini_batches 4 \ --num_epochs 1 \ --learning_rate 1e-6 \ --per_device_train_batch_size 1 \ --kl_estimator 2 \ --dataset_mixer_list ${nonreasoner_math_mix_decon} \ --dataset_mixer_list_splits train \ --dataset_mixer_eval_list hamishivi/omega-combined 4 allenai/IF_multi_constraints_upto5 4 saurabh5/rlvr_acecoder_filtered 4 hamishivi/tulu_3_rewritten_400k_string_f1_only_v2_nocode_all_filtered_qwen2_5_openthoughts2 4 hamishivi/virtuoussy_multi_subject_rlvr 4 \ --dataset_mixer_eval_list_splits train \ --max_prompt_token_length 2048 --response_length 8192 --pack_length 11264 \ --model_name_or_path ${model_name_or_path} \ --chat_template_name ${chat_template} \ --stop_strings "" \ --non_stop_penalty False \ --temperature 1.0 \ --total_episodes 1024000 \ --deepspeed_stage 3 \ --num_learners_per_node 8 \ --vllm_num_engines 56 \ --lr_scheduler_type constant \ --apply_verifiable_reward true \ --seed 1 \ --local_eval_every 50 \ --save_freq 50 \ --checkpoint_state_freq 50 \ --beaker_eval_freq 50 \ --gradient_checkpointing \ --with_tracking \ --vllm_enable_prefix_caching \ --mask_truncated_completions False \ --llm_judge_model hosted_vllm/Qwen/Qwen3-32B \ --llm_judge_timeout 600 \ --llm_judge_max_tokens 2048 \ --llm_judge_max_context_length 32768 \ --oe_eval_max_length 32768 \ --try_launch_beaker_eval_jobs_on_weka True \ --oe_eval_tasks ${general_evals_int} \ --eval_priority urgent \ --code_pass_rate_reward_threshold 0.99 \ --active_sampling \ --no_resampling_pass_rate 0.875 \