Cree1865 Tinker RL Adapter

This model card tracks the live Cree1865 synthetic-expansion training run built from Rev. E. A. Watkins' 1865 A Dictionary of the Cree Language. It is a research bootstrap artifact, not a Cree language authority, production translator, or substitute for community review.

Live Training Run

The current public run is cree1865-synthetic-expansion-v1, launched on 2026-06-26 with the Cree-specific dictionary lookup rubric.

Field	Value
W&B run	hda2wqhl
W&B project	thinking-machines-qwen3-30b
Tinker session	9d734fdb-7851-5f2f-9949-e9e574eb9a55
Base model	Qwen/Qwen3-30B-A3B-Instruct-2507
Objective	grouped rollout RL with Tinker importance_sampling
Rubric	cree
Batch size / group size	16 / 8
Planned steps	800
Approximate rollout budget	102,400 sampled completions before eval overhead
Max sampled tokens	256
Temperature	0.9
LoRA rank	32
Learning rate	4e-5
Eval / save cadence	every 20 steps
Shuffle	disabled for ordered curriculum/showcase data

Live details:

• W&B: https://wandb.ai/christian-cooper-us/thinking-machines-qwen3-30b/runs/hda2wqhl
• GitHub run note: docs/cree_synthetic_expansion_training.md
• Local log path: dakota_rl_training/outputs/cree1865_synthetic_expansion_v1

No fluency or community-validation claim is made from this run. The useful signals are per-channel reward movement, English->Cree versus Cree->English asymmetry, and orthography behavior on held-out prompts.

Source

• Source: Watkins, E. A. (1865), A Dictionary of the Cree Language
• Local PDF: CreeDictionary.pdf
• Archive master: sources/CreeDictionary_1865_cihm_41985_complete.pdf
• Structure: one book with Part I. English-Cree and Part II. Cree-English

Dataset Snapshot

Latest local full-dictionary build:

Field	Value
Extracted page JSON files	463
Raw entries	19,607
Deduplicated usable entries	19,560
Rejected incomplete entries	125
Multi-variant entries	4,049
Plain anchored Q&A rows	38,870
Core RL task rows	38,870
English->Cree core RL tasks	19,435
Cree->English core RL tasks	19,435

Synthetic expansion for the live run:

Field	Value
Expansion source	deterministic_cree_expander_v1
Variants per anchored Q&A row	6
Synthetic Q&A rows before balancing	272,090
Synthetic RL rows before balancing	272,090
Synthetic-only records	233,220
Balanced training rows	490,280
Eval probe rows	2,048
Balanced English->Cree rows	245,140
Balanced Cree->English rows	245,140
Orthography-rich output rows	184,541

The generated data lives under data/cree_goal_run_20260624_full_dictionary/training_datasets/ locally and is ignored by git because the expanded files are large. The extraction, expansion, dataset-building, and balancing code is tracked so the artifacts can be regenerated.

Reward Surface

The Prime/Verifiers environment is Cree-specific: harleycooper/cree1865-dictionary-qa version 0.1.2. It uses the Watkins dictionary QA export when repo data is present and ships a tiny packaged smoke dataset so Prime registry CI can load the environment in isolation.

The Cree task export provides Tinker-compatible fields:

• question
• prompt
• answer
• task_type
• difficulty
• info.special_chars
• info.verification_pattern
• metadata.direction

Reward channels are exact match, target containment, Cree orthography preservation, character F1, and concise answer length. There is no Dakota grammar rubric and no affix/default reward channel in the Cree verifier.

Limitations

• The source is a missionary-era 1865 dictionary with historical orthography and colonial-era framing.
• The extraction may preserve source errors, scan artifacts, and model extraction mistakes.
• The expanded prompts vary the question surface but keep dictionary-derived answers; they are not community speech data.
• Many tasks remain direct dictionary lookups, not natural dialogue.
• Cree language work should be reviewed with appropriate community and linguistic expertise.

Citation

Watkins, E. A. (1865). A Dictionary of the Cree Language, as Spoken by the Indians of the Hudson's Bay Territories. London: Society for Promoting Christian Knowledge.