HyperCLOVAX-SEED-Think-32B-mic

Slim model weights, executable usage and framework expectations in one repository.

• Contract manifest
• Transformers usage
• Transformers expectations
• Fixture provenance
• How the fixture was shrunk

Contract

HyperCLOVAX-SEED-Think-32B vision-language generation

Runnable usage and the framework expectations directory are linked above.

Slim model

Slim fixture: HyperCLOVAX-SEED-Think-32B-mic

A 9.8M-parameter fixture for the HyperCLOVAX-SEED-Think-32B MIC, covering naver-hyperclovax/HyperCLOVAX-SEED-Think-32B. It is a test artifact, not a usable model, tagged test-artifact to keep it out of candidate-model search.

• Fixture key c830339e360a. It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
• Same architecture, smaller config: HyperCLOVAXVisionV2ForConditionalGeneration, shrunk by the generic rule in sweep.py (text num_hidden_layers 2).
• Processor files copied byte for byte from naver-hyperclovax/HyperCLOVAX-SEED-Think-32B: HyperCLOVA_X_32B_Think.pdf, LICENSE, added_tokens.json, chat_template.jinja, generation_config.json, merges.txt, processor_config.json, special_tokens_map.json, tokenizer.json, tokenizer_config.json, vocab.json.
• Overfit to the contract's training step (90 steps). Random weights failed the input-sensitivity gate: attention_mask:all_ones moved the logits by only 0.00061 (threshold 0.001).
• Same tensor names as the real checkpoint. The real checkpoint has every parameter, and so does the fixture.
• Saved in bfloat16, the dtype naver-hyperclovax/HyperCLOVAX-SEED-Think-32B loads in by default.
• Finite in float32, bfloat16, and float16 (forward, loss, gradients).

Built by build_fixtures.py; build.json records the pinned sources, gate results, and file hashes. This repository includes the slim weights, usage and framework expectations; rebuild the other files with python example/build_fixtures.py --fixture HyperCLOVAX-SEED-Think-32B-mic.