Fix broken imports in 4 existing contrib models

[jimburtoft]

Description

These are fixes to existing contrib models, not new contributions.

Four contrib models crash on import due to wrong module names or case mismatches in their __init__.py or test files. All fixes are mechanical -- no logic changes.

Fixes

Model	File	Issue	Fix
Mixtral-8x7B-Instruct-v0.1	src/__init__.py	Imports from .mixtral_model but file is modeling_mixtral.py	from .modeling_mixtral import ...
OLMo-2-1124-7B	src/__init__.py	Imports from neuronx_port.modeling_olmo2 -- neuronx_port package doesn't exist	from .modeling_olmo2 import ...
Qwen3-VL-8B-Thinking	src/__init__.py	Imports from neuronx_port.modeling_qwen3_vl -- same issue	from .modeling_qwen3_vl import ...
biogpt	test/integration/test_model.py	Imports NeuronBioGPTForCausalLM / BioGPTInferenceConfig but actual class names use title case: NeuronBioGptForCausalLM / BioGptInferenceConfig	Fix all references to match actual class names

Not included

helium-1-2b also has broken imports (helium_config and helium_model modules don't exist), but this is a deeper issue -- the HeliumInferenceConfig class is referenced throughout modeling_helium.py but never defined anywhere. That model needs its config class written, not just an import path fix.

Files Changed

File	Change
contrib/models/Mixtral-8x7B-Instruct-v0.1/src/__init__.py	Fix module name
contrib/models/OLMo-2-1124-7B/src/__init__.py	Fix to relative import
contrib/models/Qwen3-VL-8B-Thinking/src/__init__.py	Fix to relative import
contrib/models/biogpt/test/integration/test_model.py	Fix class name casing

Testing

These are import-path fixes only -- no model logic is changed. Each fix corrects a ModuleNotFoundError or ImportError that prevents the contrib from being used at all.

[jimburtoft]

Validation Results

Tested on trn2.3xlarge spot instance (sa-east-1, SDK 2.28, LNC=2).

Import Validation (all 4 models)

All four models successfully import after the fix -- the classes are found and instantiated correctly:

Model	Import Test	Classes Found
Mixtral-8x7B	PASS	NeuronMixtralForCausalLM, MixtralInferenceConfig
OLMo-2-1124-7B	PASS	NeuronOlmo2ForCausalLM, Olmo2InferenceConfig, Olmo2NeuronConfig
Qwen3-VL-8B-Thinking	PASS	NeuronQwen3VLForCausalLM, Qwen3VLInferenceConfig, Qwen3VLNeuronConfig
biogpt	PASS	NeuronBioGptForCausalLM, BioGptInferenceConfig (title case confirmed)

Compilation Validation

I attempted end-to-end compile+generate on biogpt and Qwen3-VL-8B. Both hit pre-existing issues unrelated to the import fixes:

Model	Compile	Issue	Related to Import Fix?
biogpt	Compiled successfully (107.6s)	Fails at load() -- update_state_dict_for_tied_weights not implemented	No -- pre-existing missing method
Qwen3-VL-8B	Fails at config validation	Nested VL config not extracted properly by load_pretrained_config	No -- pre-existing config handling issue

These contribs have deeper issues beyond the import paths. The import fixes in this PR are correct and verified -- they resolve the ModuleNotFoundError/ImportError that prevents the models from being used at all.

[jaharsh-aws]

Do you have the github issue link for the pre-existing bugs ?