Qwen3.5-2B
A small language model in the Qwen series with 2 billion parameters, used for experimental evaluation of KV cache compression methods.
Timeline (last 90 days)
On the Qwen3.5-2B model, DCKV reduces the number of stored bytes per compressed token to more than 55 times lower than bf16, while accuracy loss on three long-context benchmarks does not exceed 1.8 percentage points.
Strands Decider 2B is based on Qwen3.5-2B plus a pointer head and LoRA, uses the Apache-2.0 license, has about 4.5GB of weights, and produces calibrated typed answers through a single forward pass.
A Reddit technical post claims that a model with only 0.8B parameters achieved a score of 42.15% on the ARC-Challenge benchmark, surpassing the untuned stock Qwen3.5-2B base model and outperforming a solution using SFT distillation with 25,000 data entries.
Comparing a specifically trained 0.8B model against an untuned stock Qwen3.5-2B base model presents an issue of unequal comparison baselines.
On the Qwen3.5-2B model, GAD-RL achieved a Micro Recall of 59.92% on CHAOS-Bench, surpassing GRPO and fixed-weight GRPO+OPD by 8.45 and 4.43 percentage points respectively
The paper's validation is mainly focused on OCR faithfulness scenarios and the Qwen3.5-2B scale model, and generalization to larger-scale models and more diverse tasks has not yet been verified
All Facts (4)
On the Qwen3.5-2B model, DCKV reduces the number of stored bytes per compressed token to more than 55 times lower than bf16, while accuracy loss on three long-context benchmarks does not exceed 1.8 percentage points.
50%UnverifiedStrands Decider 2B is based on Qwen3.5-2B plus a pointer head and LoRA, uses the Apache-2.0 license, has about 4.5GB of weights, and produces calibrated typed answers through a single forward pass.
50%UnverifiedOn the Qwen3.5-2B model, GAD-RL achieved a Micro Recall of 59.92% on CHAOS-Bench, surpassing GRPO and fixed-weight GRPO+OPD by 8.45 and 4.43 percentage points respectively
50%UnverifiedThe paper's validation is mainly focused on OCR faithfulness scenarios and the Qwen3.5-2B scale model, and generalization to larger-scale models and more diverse tasks has not yet been verified
50%