SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2601.21115 · 2026

Multi-task Code LLMs: Data Mix or Model Merge?

Stacy Patterson, Michele Merler, Mingzhi Zhu, Boris Sobolev, Rahul Krishna, Raju Pavuluri

arXiv · PDF · Open in the Atlas

Code that ran

1 of 1 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 0 failures of our instrument; 0 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2601.21115.json · badge: /api/ran/2601.21115.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
zmzfpc/Model_Merging_Data_Mixture community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
WeightPositionAnalyzer Ran
ran on a synthesized input
zmzfpc/Model_Merging_Data_Mixture/llama_factory/visual/weight_position_analysis.py
code served (permissive licence) · get_code("1014a1226c168c97")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import json
from pathlib import Path
import torch
import numpy as np

# Create temporary directories for two mock models
temp_dir = tempfile.mkdtemp()
model1_dir = Path(temp_dir) / "model1"
model2_dir = Path(temp_dir) / "model2"
output_dir = Path(temp_dir) / "output"

model1_dir.mkdir(parents=True, exist_ok=True)
model2_dir.mkdir(parents=True, exist_ok=True)

# Create mock config files
config = {
    "architectures": ["LlamaForCausalLM"],
    "model_type": "llama",
    "num_hidden_layers": 2,
    "hidden_size": 8,
    "intermediate_size": 16,
    "num_attention_heads": 2
}

with open(model1_dir / "config.json", "w") as f:
    json.dump(config, f)

with open(model2_dir / "config.json", "w") as f:
    json.dump(config, f)

# Create mock weight maps
weight_map = {
    "model.embed_tokens.weight": "model-00001-of-00001.safetensors",
    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00001.safetensors",
    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00001.safetensors",
    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00001.safetensors",
    "model.layers.0.input_layernorm.weight": "model-00001-of-00001.safetensors",
    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00001.safetensors",
    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00001.safetensors",
    "model.norm.weight": "model-00001-of-00001.safetensors",
    "lm_head.weight": "model-00001-of-00001.safetensors"
}

index1 = {"weight_map": weight_map}
index2 = {"weight_map": weight_map}

with open(model1_dir / "model.safetensors.index.json", "w") as f:
    json.dump(index1, f)

with open(model2_dir / "model.safetensors.index.json", "w") as f:
    json.dump(index2, f)

# Create mock safetensors files with actual weights
from safetensors.torch import save_file

weights1 = {
    "model.embed_tokens.weight": torch.randn(100, 8),
    "model.layers.0.self_attn.q_proj.weight": torch.randn(8, 8),
    "model.layers.0.self_attn.k_proj.weight": torch.randn(8, 8),
    "model.layers.0.mlp.gate_proj.weight": torch.randn(16, 8),
    "model.layers.0.input_layernorm.weight": torch.randn(8),
    "model.layers.1.self_attn.q_proj.weight": torch.randn(8, 8),
    "model.layers.1.mlp.gate_proj.weight": torch.randn(16, 8),
    "model.norm.weight": torch.randn(8),
    "lm_head.weight": torch.randn(100, 8)
}

weights2 = {
    "model.embed_tokens.weight": torch.randn(100, 8),
    "model.layers.0.self_attn.q_proj.weight": torch.randn(8, 8),
    "model.layers.0.self_attn.k_proj.weight": torch.randn(8, 8),
    "model.layers.0.mlp.gate_proj.weight": torch.randn(16, 8),
    "model.layers.0.input_layernorm.weight": torch.randn(8),
    "model.layers.1.self_attn.q_proj.weight": torch.randn(8, 8),
    "model.layers.1.mlp.gate_proj.weight": torch.randn(16, 8),
    "model.norm.weight": torch.randn(8),
    "lm_head.weight": torch.randn(100, 8)
}

save_file(weights1, model1_dir / "model-00001-of-00001.safetensors")
save_file(weights2, model2_dir / "model-00001-of-00001.safetensors")

INIT_ARGS = {
    "model1_path": str(model1_dir),
    "model2_path": str(model2_dir),
    "output_dir": str(output_dir)
}

METHOD = "analyze_position_distributions"

ARGS = {
    "weights1": weights1,
    "weights2": weights2
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
matplotlib, numpy, pandas, safetensors, scipy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)

Also present in other papers' code

Of 1 samples lifted from this paper's linked repositories, 1 carry a function identity and 0 do not. Of the 1 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 1 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Recent research advocates deploying smaller, specialized code LLMs in agentic frameworks alongside frontier models, sparking interest in efficient strategies for multi-task learning that balance performance, constraints, and costs. We compare two approaches for creating small, multi-task code LLMs: data mixing versus model merging. We conduct extensive experiments across two model families (Qwen Coder and DeepSeek Coder) at two scales (2B and 7B parameters), fine-tuning them for code generation and code summarization tasks. Our evaluation on HumanEval, MBPP, and CodeXGlue benchmarks reveals that model merging achieves the best overall performance at larger scale across model families, retaining 96% of specialized model performance on code generation tasks while maintaining summarization capabilities. Notably, merged models can even surpass individually fine-tuned models, with our best configuration of Qwen Coder 2.5 7B model achieving 92.7% Pass@1 on HumanEval compared to 90.9% for its task-specific fine-tuned equivalent. At a smaller scale we find instead data mixing to be a preferred strategy.We further introduce a weight analysis technique to understand how different tasks affect model parameters and their implications for merging strategies. The results suggest that careful merging and mixing strategies can effectively combine task-specific capabilities without significant performance degradation, making them ideal for resource-constrained deployment scenarios. Our code can be found at https://github.com/zmzfpc/Model_Merging_Data_Mixture. • Computing methodologies → Modeling methodologies.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2601.21115")
get_code_for_paper("2601.21115")
have("2601.21115")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2601.21115.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2601.21115.svg)](https://syntology.ai/paper/2601.21115)

Connect an agent — have() is free.