🔬Modal Research Exchange

{
  "base_model": "Qwen/Qwen3.5-4B",
  "baseline": "dimuon-lite (DiMuon with Polar Express), beta1=0.9, grad_clip_norm=0.0. Same model, rank, data order, seed, renderer, batch size, steps and LR grid. The best-LR run of each arm is compared.",
  "batch_size": 64,
  "budget": {
    "gpu_hours": 1.8,
    "max_wall_minutes": 120
  },
  "command": "python - <<'EOF'\n# Custom part: the TriMuon config exists only on the fork; the baseline uses tinker.types.DimuonOptimizerConfig() / DimuonParams.\nimport os, tinker\nfrom tinker_cookbook.recipes.chat_sl import chat_datasets  # renderer + no_robots builder\nsc = tinker.ServiceClient(base_url=os.environ[\"TINKER_BASE_URL\"])  # fork URL for the TriMuon arm\nARM = os.environ.get(\"ARM\", \"pe\"); LR = float(os.environ.get(\"LR\", \"1e-4\"))\nopt_cfg = tinker.types.DimuonOptimizerConfig() if ARM == \"pe\" else tinker.types.TrimuonOptimizerConfig()  # Trimuon* added by the fork\ntc = sc.create_lora_training_client(base_model=\"Qwen/Qwen3.5-4B\", rank=32, optimizer=opt_cfg)\ntrain, test = chat_datasets.no_robots_batches(tc.get_tokenizer(), batch_size=64, max_length=2048, seed=0)\nfor step, batch in zip(range(80), train):\n    tc.forward_backward(batch, loss_fn=\"cross_entropy\").result()\n    P = tinker.types.DimuonParams if ARM == \"pe\" else tinker.types.TrimuonParams\n    tc.optim_step(P(learning_rate=LR, beta1=0.9, grad_clip_norm=0.0)).result()\nprint(\"eval_nll\", sum(tc.forward(b, loss_fn=\"cross_entropy\").result().metrics[\"loss:mean\"] for b in test) / len(test))\ntc.save_weights_for_sampler(name=f\"rx-trimuon-{ARM}-{LR}\")\nEOF",
  "dataset": {
    "hf_path": "HuggingFaceH4/no_robots",
    "n_examples": 5120,
    "name": "no_robots",
    "split": "train (eval on test)"
  },
  "eval": {
    "benchmark": "no_robots test split",
    "metric": "mean per-token NLL on assistant tokens",
    "n_examples": 500
  },
  "hypothesis": "With a Spindle fork that adds TriMuon, LoRA SFT on Qwen3.5-4B reaches held-out no_robots loss at least 0.01 nats lower than dimuon-lite (Polar Express) at its best LR, after the same 80 steps.",
  "lora_rank": 32,
  "max_seq_len": 2048,
  "method": "Fork modal-projects/spindle and add TriMuon as an optimizer option. It keeps DiMuon's recovered momentum and preconditioning rules. It replaces only the Polar Express polar step with two outer Polar Switch passes, then four cubic Newton-Schulz refinements folded into the rank x rank multiplier. Use the FP16 polar core and TF32 disabled, as in aaron's benchmark. Deploy the fork as a throwaway copy. Train rank-32 LoRA on Qwen/Qwen3.5-4B with a short tinker SDK SFT loop on no_robots. The loop uses the cookbook renderer for the model, forward_backward with cross_entropy, then optim_step. The treatment arm runs on the fork with TriMuon. The baseline arm runs dimuon-lite on the same fork (or shared Spindle) with identical code. Each arm sweeps LR over {5e-5, 1e-4, 2e-4}: 6 runs of 80 steps each, same data order and seed. Each run evaluates held-out loss at step 80 and logs per-step optimizer wall time. Qwen3.5-4B is a scaled-down stand-in for the source's Qwen3-8B rank-32 workload. Polar Switch and AdamW arms are follow-up specs, not part of this decision.",
  "optimizer": {
    "lr": 0.0001,
    "name": "trimuon (fork)"
  },
  "steps": 80,
  "success_criterion": "Validated if best-LR TriMuon eval NLL is at least 0.01 nats below best-LR dimuon-lite eval NLL, and TriMuon's median optimizer step time is not higher. Failed if the gap is under 0.01 nats or reversed. A rerun of the best PE config with a second seed is optional, if budget allows, to check that the 0.01 margin exceeds seed noise.",
  "title": "TriMuon vs dimuon-lite (Polar Express) for LoRA SFT on no_robots",
  "training_type": "sft"
}

• script · probe_raw.py
• metrics · probe_raw.log
• code_revision · code-revision.txt

1. Nov 3, 23:25 UTC · failed · Preflight on shared Spindle returned HTTP 400 at create_model for both arms ('optimizer family dimuon is not supported', 'optimizer family trimuon is not supported'), so both arms would need a fork. …

adaptation_manifests

[]

correctness_preflight

{'required': True, 'status': 'unverified', 'paths': [], 'missing': ['dimuon_lite_pe: the spec pins no baseline or reference revision and no tolerances. Shared Spindle main (afd2f22) rejects dimuon, so it would have to be built from withdrawn modal-projects/spindle PR #194 / modal-projects/miles PR #17 head 228b59d9', "trimuon: modal-projects/miles PR #27 head 45937c3b is an empty 'Withdrawn' commit on top of PR #17, so the TriMuon code is gone. The spec pins no reference revision for the 2x Polar Switch + 4 cubic NS refinement or the FP16 core, and gives no tolerance"], 'requested_revision': 'Pin (a) the dimuon-lite reference commit (e.g. miles 228b59d9), (b) Polar Switch 6f382f1e PS3_55 coefficients and an exact TriMuon reference (fp64 torch) with the refinement and fold definition, (c) atol/rtol/min_cosine for update parity and LR-scaling tolerance, and (d) a Miles trainer image revision for Qwen3.5-4B.'}

correctness_validation

{'status': 'unverified', 'required': True, 'checks': [], 'reason': 'Correctness preflight was not recorded.', 'evidence_source': 'executor report; remote execution is not attested', 'applicability': 'unknown', 'reported_verdict': 'inconclusive'}

feasibility

{'status': 'unknown', 'problems': ['Measured feasibility not recorded'], 'wall_seconds': None, 'gpu_hours': None, 'known_gpu_hours': 0, 'phases': {}, 'record': None, 'recorded_at': 1793748094.8738408}

feasibility_required

True

initialization_provenance

{'version': 1, 'arms': {}, 'note': 'not created; no training client was successfully created for either arm'}

last_reply

head 45937c3b is an empty 'Withdrawn' commit on top of PR #17, so the TriMuon code is gone. The spec pins no reference revision for the 2x Polar Switch + 4 cubic NS refinement or the FP16 core, and gives no tolerance"], "requested_revision": "Pin (a) the dimuon-lite reference commit (e.g. miles 228b59d9), (b) Polar Switch 6f382f1e PS3_55 coefficients and an exact TriMuon reference (fp64 torch) with the refinement and fold definition, (c) atol/rtol/min_cosine for update parity and LR-scaling tolerance, and (d) a Miles trainer image revision for Qwen3.5-4B."}, "initialization_provenance": {"version": 1, "arms": {}, "note": "not created; no training client was successfully created for either arm"}, "activation_calibration": {"values": {}, "note": "no pilot ran; both optimizer families were rejected before model creation"}, "feasibility": null, "loss_eval": null, "adaptation_manifests": [], "evidence": {"version": 1, "files": [{"name": "probe_raw.py", "kind": "script", "content": "# Raw HTTP probe of shared Spindle create_model with each arm's optimizer_config (Qwen/Qwen3.5-4B, LoRA rank 32); credentials read from env vars; session finished afterwards. sha256 1bae750cc5f3c601d49e8df6acc005d7f8c9ab39efafb7e4a77528ca7a7b366e"}, {"name": "probe_raw.log", "kind": "metrics", "content": "create_session 200\ncreate_model {'type': 'dimuon', 'version': 1} 400 {\"error\":\"invalid_request\",\"message\":\"optimizer family dimuon is not supported\"}\ncreate_model {'type': 'trimuon'} 400 {\"error\":\"invalid_request\",\"message\":\"optimizer family trimuon is not supported\"}\nfinish_session 200 null"}, {"name": "code-revision.txt", "kind": "code_revision", "content": "spindle main afd2f22 (rejects non-adamw in control_plane/http.py create_model); spindle PR#194 head e23308c 'Withdrawn' (dimuon only, no trimuon/polar code); miles PR#17 head 228b59d9 (dimuon-lite); miles PR#27 head 45937c3b 'Withdrawn' (empty commit, TriMuon removed)"}], "checkpoints": [], "per_arm_metrics": {}}} ```

launch

{'experiment_id': 'exp_ff293ca443', 'context_status': 'legacy_unknown', 'superseded': False, 'mode': 'redo', 'purpose': "Repeat the selected saved experiment version to check reproducibility. This repetition tests that version's hypothesis and methods."}

progress_updated_at

1793748279.6005318

reported_outcome_reason

incomplete_execution

reported_verdict

inconclusive

restoration_validation

{'status': 'unverified', 'required': False, 'checks': [], 'reason': 'Initialization and optimizer evidence was not required by this saved spec.', 'applicability': 'unknown'}

started_at

1793748032.9418862