CloudGoat / Mephisto-4B-v2.1

Mephisto-4B-v2.1

Abstract

Mephisto-4B-v2.1 is a 4B-parameter agentic language model constructed via multi-stage model merging using mergekit. The model integrates reasoning, coding, and agentic capabilities into a single 4B checkpoint based on the Qwen3.5-4B architecture, targeting performance exceeding Mephisto-4B-v2 on Japanese and agentic benchmarks while maintaining deployment feasibility on consumer hardware (RTX 3060 12GB).

Methodology

Merge Pipeline

The model is constructed through a three-stage merging pipeline:

Stage 1: Reasoning-Code Fusion (NuSLERP)

Normalized spherical linear interpolation (NuSLERP) on task vectors derived from a common base model (Qwen3.5-4B). This stage fuses a reasoning-distilled model (Opus-reasoning distillation) with a code-specialized model.

Configuration:

merge_method: nuslerp
base_model: Qwen/Qwen3.5-4B
models:
  - model: Jackrong/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled-v2
    parameters: {weight: 1.0}
  - model: Jackrong/Qwopus3.5-4B-Coder
    parameters: {weight: 0.7}
parameters:
  nuslerp_flatten: true
  nuslerp_row_wise: false
dtype: bfloat16

Stage 2: Agent Foundation (DARE-TIES 3-way)

DARE-TIES merging with density-based pruning and sign consensus. This stage integrates three complementary agentic models into a unified foundation.

Configuration:

merge_method: dare_ties
base_model: Qwen/Qwen3.5-4B
models:
  - model: BAAI/AREX-Turbo
    parameters: {weight: 1.0, density: 0.5}
  - model: InternScience/Agents-A1-4B
    parameters: {weight: 1.0, density: 0.5}
  - model: Jackrong/Qwopus3.5-4B-v3
    parameters: {weight: 0.6, density: 0.5}
parameters:
  lambda: 0.8
  density: 0.5
dtype: bfloat16

Stage 3: Final Integration (FrankenMerge / Passthrough)

Layer-wise model selection (FrankenMerge) via Passthrough merge. This stage structurally allocates layers by functional specialization:

Configuration:

merge_method: passthrough
slices:
  - sources:
      - model: Qwen/Qwen3.5-4B
        layer_range: [0, 16]
  - sources:
      - model: merged/s2
        layer_range: [16, 24]
  - sources:
      - model: merged/s1
        layer_range: [24, 32]

Source Models

All source models are Apache-2.0 licensed, enabling commercial use of the merged artifact.

Benchmarks (In Progress)

Benchmark evaluation is currently in progress. Target thresholds are based on Mephisto-4B-v2 baseline and merge methodology expectations.

Quantization

Note: The --no-mtp flag is required to exclude Multi-Token Prediction (MTP) modules present in Qwen3.5-series models, which are incompatible with llama.cpp inference.

Usage

HF Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("Mephisto-4B-v2.1", torch_dtype="bfloat16", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("Mephisto-4B-v2.1")

llama.cpp (Recommended)

./llama-cli -m Mephisto-4B-v2.1_q8_0.gguf -ngl 99 -c 4096 -p "ユーザー: こんにちは
アシスタント: "

Hardware Requirements

License

This model inherits licenses from its source models. All source models are Apache-2.0 licensed, permitting commercial use. Users should verify individual source model licenses for their specific use cases.

Citation

If you use this model in your research, please cite:

@misc{mephisto-4b-v2.1,
  title={Mephisto-4B-v2.1: Multi-Stage Merged Agentic Model},
  author={CloudGoat},
  year={2024},
  note={Constructed via mergekit multi-stage merging: NuSLERP + DARE-TIES + FrankenMerge}
}

---

Model Card Version: 1.0
Last Updated: 2024-08-05
Merge Tool: mergekit (NuSLERP + DARE-TIES + Passthrough)
Base Architecture: Qwen3.5-4B (32 layers, 2560 hidden dim, 16/4 attention heads)