jacob-valdez/tensorcode-planner-hotpot-001

TensorCode Planner support relevance prototype

Owned encoder, shared differentiable workspace and candidate scoring head; the manifest identifies inherited foundation weights when used. This checkpoint ranks supplied HotpotQA documents using human supporting-fact annotations. It is not a general cognitive agent. Planner scores are relevance proxies, not measured plan utility. Investigator target policy is in the manifest; foundation runs supervise all supporting titles. Newly built vocabulary and gradients use training questions only; foundation tokenizer assets are inherited; held-out official validation IDs are fixed before fitting. The final epoch is saved without validation-based selection. Inherited foundation models, if used, are pinned in the manifest.

Load with from tensorcode.tools.planner import Planner then Planner.from_pretrained(path_or_repo). Inputs contain question/hypotheses for Investigator or goal/plans for Planner; each candidate has id/text. See the TensorCode example for the complete schema.

Measured results

{
  "before": {
    "loss": 0.19878229580353945,
    "support_hit_at_1": 0.28125,
    "support_recall_at_2": 0.25
  },
  "after": {
    "loss": 0.14763107878388837,
    "support_hit_at_1": 0.5546875,
    "support_recall_at_2": 0.46484375
  },
  "zero_workspace": {
    "loss": 0.1479919097328093,
    "support_hit_at_1": 0.5234375,
    "support_recall_at_2": 0.4765625
  },
  "reloaded": {
    "loss": 0.14763107878388837,
    "support_hit_at_1": 0.5546875,
    "support_recall_at_2": 0.46484375
  },
  "training_loss": [
    0.15151852124836296,
    0.14454737049527466,
    0.14483398525044322,
    0.14204114448511973,
    0.14295739110093564,
    0.14226798515301198,
    0.14272308524232358,
    0.14249018975533545
  ],
  "development_metrics": [
    {
      "loss": 0.1484826555242762,
      "support_hit_at_1": 0.5390625,
      "support_recall_at_2": 0.44140625
    },
    {
      "loss": 0.14597285696072504,
      "support_hit_at_1": 0.5859375,
      "support_recall_at_2": 0.45703125
    },
    {
      "loss": 0.14712385350139812,
      "support_hit_at_1": 0.5625,
      "support_recall_at_2": 0.41796875
    },
    {
      "loss": 0.14488179853651673,
      "support_hit_at_1": 0.6171875,
      "support_recall_at_2": 0.43359375
    },
    {
      "loss": 0.14909204523428343,
      "support_hit_at_1": 0.609375,
      "support_recall_at_2": 0.43359375
    },
    {
      "loss": 0.14556936410372145,
      "support_hit_at_1": 0.59375,
      "support_recall_at_2": 0.44140625
    },
    {
      "loss": 0.14576854620827362,
      "support_hit_at_1": 0.5546875,
      "support_recall_at_2": 0.41796875
    },
    {
      "loss": 0.14373427545069717,
      "support_hit_at_1": 0.5859375,
      "support_recall_at_2": 0.43359375
    }
  ]
}

Data, split IDs, hashes, hyperparameters and limitations are recorded in training-manifest.json. Attention is learned routing, not proof of factual support. Scores are uncalibrated.

Authored lexical comparison

{
  "method": "count unique casefold alphanumeric question token overlap with full candidate passage; ties candidate order",
  "support_hit_at_1": 0.53125,
  "support_recall_at_2": 0.4296875
}

The zero-workspace ablation does not demonstrate a consistent workspace benefit across metrics on this held-out subset. This release establishes a narrow support-ranking model, not general reasoning or planning.