Skip to main content
When something is wrong with a Miles run, the question is almost always: rollout or training? Once you’ve isolated which side is misbehaving, the rest is a normal debugging session.

​
Aligning precision

The single most common class of bug. Walk through these checks before anything else.

​
First training step

​
Is the rollout coherent?

If the very first rollout is gibberish:
  • Parameters didn’t load. Check Megatron’s logs — there should be a clear “loaded checkpoint from …” line. If absent, fix --load / --ref-load.
  • Parameter mapping is wrong. When pp_size > 1, the second-stage layer IDs are a common offset bug. Save all parameters in load_weights of the corresponding model in SGLang and verify they match the loaded checkpoint exactly.
  • SGLang dropped buffers. Some special buffers can be released during the parameter release process. Check if they’re re-loaded correctly after weight sync.
  • Pretrained vs. instruct. Try the instruct version of the same architecture. If that works, your pretrained model + chat template combination is wrong.

​
Are log_probs and ref_log_probs exactly equal in step 1?

They should be: at step 1 the actor and reference are the same weights. If KL > 0:
  • Non-deterministic kernels. Some Transformer Engine versions need --attention-backend flash to force deterministic Flash Attention under context parallel.
  • Slightly off values (KL < 1e-4). Acceptable — kernel-level numerical jitter.
  • Large values (KL > 1). Configuration error. Re-check parallelism and precision.
  • Slightly elevated logp on instruct (~0.8 per token). Almost always a chat-template mismatch — your prompts don’t match the format the model was trained on. Run the chat template verifier.

​
Is grad_norm reasonable?

Step 1 with num_steps_per_rollout=1 should produce a tiny gradient. If it doesn’t:
  • Megatron / TE bug. MoE in particular requires --moe-permute-fusion. Check release notes for known fixes.
  • Reward signal is broken. Confirm rewards are computed from the right key (often a --label-key typo).

​
Second training step

If step 2 OOMs in colocate mode, Megatron’s offload→reload cycle is hitting the SGLang side. Lower --sglang-mem-fraction-static to 0.7 (or 0.6).

​
Separate train/rollout debugging

Miles ships flags to isolate one side:
FlagWhat
--debug-rollout-onlyDon’t init Megatron — only spin up SGLang.
--debug-train-onlyDon’t init SGLang — only spin up Megatron.
--save-debug-rollout-data /path/data_{rollout_id}.ptPickle every rollout to disk.
--load-debug-rollout-data /path/data_{rollout_id}.ptReplay rollouts from disk; auto-sets --debug-train-only.
A typical workflow:
  1. Run with --debug-rollout-only --save-debug-rollout-data to capture a few well-formed rollouts.
  2. Switch to --debug-train-only --load-debug-rollout-data to iterate on training changes (parallelism, optimizer, custom loss) with fixed inputs. Removes rollout randomness from your bisect.
This is the single most useful pattern in the Miles workflow. Use it.

​
Determinism for bisecting

When you need to A/B test a code change, bit-wise reproducibility is your friend. See the Reproducibility recipe for the exact flag set and env vars. The 25% throughput cost is worth it during development.

​
Common kernel pitfalls

SymptomLikely culprit
Garbled rollout, parameters loaded fineChat template mismatch, or buffer drop in SGLang
KL ≠ 0 in step 1Non-det fused attention; force --attention-backend flash
MoE training collapses after ~50 stepsR3 not on, or routing not preserved
Gradient NaN/InfBad chat template, or activation overflow in FP8
illegal memory access in SGLangOOM in disguise — lower --sglang-mem-fraction-static
JSONDecodeError from inductorCache corrupt; set TORCHINDUCTOR_FORCE_DISABLE_CACHES=1
NCCL hang during weight syncNCCL_TIMEOUT=900; check NCCL_DEBUG=INFO

​
Reading logs

Where to look:
ComponentPath
Trainer stdoutwherever you redirected ray job submit
SGLang/tmp/sglang/*.log (or --sglang-log-dir)
Ray workers~/.ray/session_latest/logs/worker-*.{out,err}
NCCLNCCL_DEBUG=INFO NCCL_DEBUG_FILE=/tmp/nccl_%h_%p.log
grep for these signal phrases:
  • Loaded checkpoint from — Megatron load OK.
  • weight_sync — P2P / broadcast events.
  • Continuous async rollout worker started — async worker alive.
  • WorkerCrashed / RaySystemError — ranks died.
  • forward_logp_max_diff — FP8 train/inference alignment.

​
When all else fails

  • Drop to a tiny model (Qwen2.5-0.5B) on a known-good recipe (Reproducibility) to isolate framework vs. model.
  • git bisect between a known-working commit and HEAD. Determinism makes this trustworthy.
  • Open a GitHub issue with: launch script, pip freeze, the first 200 lines of trainer stdout, and a short description.

​
Useful debugging knobs

--debug-determinism                      # log per-step hashes
--debug-weight-sync 1                    # verbose P2P transfer logs
--debug-rollout-print-every 1            # dump every Nth rollout to stdout
NCCL_DEBUG=INFO
NCCL_DEBUG_SUBSYS=COLL,P2P
RAY_DEDUP_LOGS=0