internvl3-5-8b-text-only

InternVL3.5-8B Text-Only

This is a text-only extraction of OpenGVLab/InternVL3_5-8B. It keeps the Qwen3 language model weights and tokenizer assets, and removes the vision/multimodal components.

This repository is not an official OpenGVLab release.

Conversion

• Source model: OpenGVLab/InternVL3_5-8B
• Source revision: 9bb6a56ad9cc69db95e2d4eeb15a52bbcac4ef79
• Source architecture: InternVLChatModel
• Target model type: qwen3
• Target architecture: Qwen3ForCausalLM
• Kept tensors: 399
• Dropped tensors: 346
• Dropped components: vision_model.* and InternVL bridge mlp1.*
• Weight format: sharded safetensors
• Weight files: 4 shards at repository root

The original VLM config stores the text backbone under llm_config and the weights under language_model.*. The extracted checkpoint strips that prefix and writes a standalone Qwen3 CausalLM config for AutoModelForCausalLM.

Validation

Validated locally with Transformers:

• AutoConfig.from_pretrained(...) loads as Qwen3Config
• AutoTokenizer.from_pretrained(...) loads successfully
• AutoModelForCausalLM.from_pretrained(..., torch_dtype="auto", low_cpu_mem_usage=True) loads as Qwen3ForCausalLM
• A tiny forward pass succeeds and returns logits with shape (1, 1, 151936)
• The sharded safetensors index contains 399 tensors
• No vision/projector/bridge/language wrapper tensor names remain in the sharded weight index
• Shard headers match model.safetensors.index.json

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "sasa2000/internvl3-5-8b-text-only"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

inputs = tokenizer("Explain why the sky looks blue.", return_tensors="pt").to(model.device)
with torch.no_grad():
    output_ids = model.generate(**inputs, max_new_tokens=128)

print(tokenizer.decode(output_ids[0], skip_special_tokens=True))

Limitations

This checkpoint is text-only. Image and video inputs are not supported because the vision encoder, InternVL bridge, processor assets, and related preprocessing files were removed.

Please review the upstream model card and license for the original model's intended use, limitations, and terms.