Add single-node budget tutorials (slime/ms_swift/miles) + cost estimation#2
Closed
devin-ai-integration[bot] wants to merge 4 commits into
Closed
Conversation
- Expand GPUType to include T4, L4, A10, L40S, A100 (all variants) - Add modal_training_gym/common/cost.py with GPU price table + estimate_cost() - Add 'Est. Cost/hr' column to tutorials/README.md (auto-generated from gpu_type + n_gpus metadata fields) - New tutorials: - sft/002_ms_swift_qwen3_0_6b: Qwen3-0.6B LoRA SFT on 1×A10 (~$1.10/hr) - rl/005_slime_single_gpu: Qwen3-0.6B GRPO on 2×A100 (~$5.00/hr) - rl/005_miles_intro: Qwen3-0.6B GRPO with Miles on 8×A100 (~$20.00/hr) - Add cost callouts to existing tutorial narratives (001_slime_intro, 001_ms_swift) - Add gpu_type/n_gpus metadata to all existing tutorials Co-Authored-By: Joy <joyliu.q@gmail.com>
Contributor
Author
🤖 Devin AI EngineerI'll be helping with this pull request! Here's what you should know: ✅ I will automatically:
⚙️ Control Options:
|
Every tutorial now has a 'Cost estimate' section with a runnable code cell that imports from modal_training_gym.common.cost and prints the estimated cost for the smoke run and a 1-hour full run. Users can adjust estimated_minutes before launching. Co-Authored-By: Joy <joyliu.q@gmail.com>
Remove T4, L4, A10, L40S, B300 from supported GPU list. Retarget the ms_swift budget tutorial from A10 to A100-40GB. Update cost tables in cost.py and generate_tutorial.py to match. Co-Authored-By: Joy <joyliu.q@gmail.com>
Co-Authored-By: Joy <joyliu.q@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds three new single-node, budget-GPU tutorials using Qwen3-0.6B, plus cost estimation infrastructure so users can see what each tutorial costs before launching.
New tutorials
sft/002_ms_swift_qwen3_0_6brl/005_slime_single_gpurl/005_miles_introAll three use Qwen3-0.6B and target cheaper GPU tiers (A100, A100-40GB) instead of H100. Each includes cost callouts in the narrative, smoke-run defaults (
train_iters=5, small batch, tiny dataset slice), and retargets the model's preset via a lightweight subclass.Cost estimation
modal_training_gym/common/cost.py—GPU_HOURLY_PRICESdict +estimate_cost()/format_cost_range()utilities.cost.pyand prints estimated costs for smoke runs and full runs. Users can adjustestimated_minutesbefore launching.gpu_type+n_gpusmetadata fields in each tutorial'sTUTORIAL_METADATA.Infrastructure
GPUTypenarrowed to supported tiers:A100,A100-40GB,A100-80GB,H100,H200,B200.Checklist
latestpython_versionfor the base image, if it is used~=x.y.zor==x.yversion < 1are pinned to patch version,==0.y.zLink to Devin session: https://app.devin.ai/sessions/21d21063333249d292cbbab7064ff905
Requested by: @joyliu-q