Laya fine-tuned for four news questions

README.md

@@ -0,0 +1,109 @@
+---
+license: apache-2.0
+base_model: convaiinnovations/laya
+language:
+- multilingual
+tags:
+- laya
+- system-one
+- news
+- classification
+- rlcd
+pipeline_tag: text-classification
+---
+
+# laya-news-decisions
+
+A fine-tuned [Laya](https://huggingface.co/convaiinnovations/laya) (multilingual
+variant, 322M parameters) that answers four typed questions about a news article
+in one forward pass:
+
+| key | type | question |
+|---|---|---|
+| `topic` | choice, 10 classes | which category does this news report belong to |
+| `impact` | score, 5 levels | how far do the consequences reach, from one person to global |
+| `violence` | yes/no | does the article describe physical violence |
+| `civilian_harm` | score, 5 levels | how severe are the consequences for civilians |
+
+It was trained with [laya-rlcd-training](https://huggingface.co/InfinimindCreations/laya-rlcd-training),
+the open training loop we wrote for Laya. This model is what that loop produces
+on a real corpus.
+
+## Use
+
+```python
+import json, laya
+from huggingface_hub import hf_hub_download
+from safetensors.torch import load_file
+
+repo = "InfinimindCreations/laya-news-decisions"
+agent = laya.load("convaiinnovations/laya", device="cpu", subfolder="multilingual")
+agent.model.load_state_dict(load_file(hf_hub_download(repo, "model.safetensors")), strict=False)
+questions = json.load(open(hf_hub_download(repo, "questions.json")))
+
+text = "Floods in Bangladesh displace 40,000 families"
+print(agent.predict({"headline": text}, questions)["answers"])
+```
+
+The state key is `headline` because that is the field the model was trained on,
+but we fed it the article body (first 2,800 characters) during training, and
+that is what it expects. Titles alone work, less well.
+
+**Use `questions.json` exactly as it is.** The model is sensitive to the wording
+of the options it was trained with, not to the instruction sentence. Two option
+descriptions start with the literal text `NEU v2.`, a leftover editing note from
+our label schema. It is part of the trained wording; removing it changes the
+answers. Renaming the question keys does not (checked on four texts, identical
+output).
+
+## Training
+
+- About 92,000 news articles from 2026, many languages, labelled on the full text
+  by DeepSeek v4.1 Flash.
+- One pass, full fine-tune on one A100, batch 16, G=32, lr 1e-5 with cosine
+  decay. Roughly 11 USD.
+- Only four questions on purpose: a seven-question version of the same model was
+  worse on four of six shared questions. Every extra question costs the others.
+
+## Evaluation
+
+176 gold judgements, each made by three independent LLM annotators and kept by
+majority vote. Accuracy is exact match; for the ordinal questions `±1` is the
+share within one level and `ρ` the rank correlation.
+
+| question | selected checkpoint | end of run | notes |
+|---|---|---|---|
+| topic | 0.784 | 0.756 | majority class 0.148 |
+| impact | 0.614 | 0.557 | ±1 0.977 · ρ 0.775 |
+| violence | 0.943 | 0.966 | AUC 0.977 |
+| civilian harm | 0.659 | 0.642 | ±1 0.943 · ρ 0.707 |
+
+**Read the first column with care.** The released checkpoint was selected as the
+best of 14 evaluations on this same gold set, so its numbers are the optimistic
+end. The second column is the last checkpoint of the run and is closer to what
+you should expect on unseen data. Adjacent evaluations moved by up to 4.5 points on
+topic, which is eight articles out of 176.
+
+For scale, the labelling models on the same 176 articles and the same topic
+question: DeepSeek v4.1 Flash 0.835, Gemini 3 Flash 0.812, Claude Haiku 4.5
+0.778. The fine-tuned 322M encoder sits in that range at a fraction of the cost
+and runs locally.
+
+Where it is weak: `politics`. It gets half of those right, while its teacher got
+three quarters. Better labels did not fix it (we tried, from scratch and as a
+continuation, both worse), so it is capacity or data volume, not label quality.
+
+## Limits
+
+- The gold set is small and curated. It over-represents high-impact articles
+  compared to a real news stream, so precision in the field will be lower than
+  these numbers suggest.
+- Topic categories overlap in the real world (is a sanctions package politics or
+  finance?). Even the large labelling models disagree with each other on about
+  one article in six.
+- The 0.5 cut-off is not a decision boundary for `violence`. Fit your threshold on
+  your own data.
+
+## License
+
+Apache 2.0, following Laya itself.

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56b22b357120265cd8a2611638ae25c4ffbb853cf80ba744d9d49df766fd088a
+size 1287653720

questions.json

@@ -0,0 +1,48 @@
+{
+  "topic": {
+    "type": "choice",
+    "instructions": "Which category does this news report belong to?",
+    "criteria": {
+      "conflict": "armed conflict, war, military operations, attacks, strikes on targets",
+      "security": "internal security, police, crime, terrorism response, cybersecurity, intelligence services. NOT everything that merely sounds dangerous.",
+      "politics": "government, elections, diplomacy, legislation, parties, court rulings on political matters",
+      "finance": "economy, markets, companies, currency, trade, business crime",
+      "technology": "technology, IT, AI, science, space",
+      "health": "medicine, disease, healthcare system, epidemics",
+      "humanitarian": "displacement, hunger, aid operations, human-rights situation, refugee and civilian suffering in a crisis",
+      "disaster": "NEU v2. earthquakes, floods, storms with damage, fires, industrial and traffic accidents with casualties. The EVENT itself; the aid response to it is humanitarian.",
+      "infrastructure": "NEU v2. power supply, transport networks, water, telecoms, pipelines: outages, failures, maintenance, expansion. Drought as a supply problem belongs here, drought as a famine belongs to humanitarian.",
+      "off-topic": "sport, entertainment, celebrities, lifestyle, plain weather forecasts without an event, and everything that is not a real article"
+    }
+  },
+  "impact": {
+    "type": "score",
+    "instructions": "How far do the consequences of this event reach?",
+    "criteria": [
+      "single person or single place",
+      "one city or one organisation",
+      "one country or one sector",
+      "several countries or a whole industry",
+      "global or systemic"
+    ]
+  },
+  "violence": {
+    "type": "noul",
+    "instructions": "Does this news report describe physical violence against people?",
+    "criteria": {
+      "true": "killing, wounding, attack, assault, bombardment against people",
+      "false": "everything else, including threats, weapons deals and verbal attacks"
+    }
+  },
+  "civilian_harm": {
+    "type": "score",
+    "instructions": "How severely are civilians affected?",
+    "criteria": [
+      "civilians not affected",
+      "civilians affected indirectly (restrictions, supply)",
+      "few civilians injured or displaced",
+      "many civilians affected, deaths",
+      "mass casualties among civilians"
+    ]
+  }
+}