Model · wizardcoder-15b-v1-0

WizardCoder-15B-V1.0

Developer: Ziyang Luo (HKBU), Can Xu, Pu Zhao, Qingfeng Sun, Xiubo Geng, Wenxiang Hu, Chongyang Tao, Jing Ma, Qingwei Lin, Daxin Jiang (Microsoft) — WizardLM team; HF org WizardLMTeam
Availability: available · checked 2026-09-24 · source

Raw record: /data/models/wizardcoder-15b-v1-0.json

Fields

id

wizardcoder-15b-v1-0

identifiers

huggingface

WizardLMTeam/WizardCoder-15B-V1.0

developer

Ziyang Luo (HKBU), Can Xu, Pu Zhao, Qingfeng Sun, Xiubo Geng, Wenxiang Hu, Chongyang Tao, Jing Ma, Qingwei Lin, Daxin Jiang (Microsoft) — WizardLM team; HF org WizardLMTeam

release_date

2023-06-14partial · source

note:  HF repo creation date (2023-06-14). The WizardCoder paper (arXiv 2306.08568) was first posted in June 2023 (v2 read: 27 May 2025); card date of release not stated.

weights_status

open

availability

available · checked 2026-09-24 · source

license

bigscience-openrail-m (as written on the card; StarCoder itself is BigCode OpenRAIL-M)partial · source

note:  Card metadata: `license: bigscience-openrail-m`. The base StarCoder is under bigcode-openrail-m (the BigScience RAIL family is a different license line). Discrepancy not resolved; reviewer check.

architecture

family

decoder_only

note

family read from config 'architectures': ['GPTBigCodeForCausalLM']. Config vocab_size 49,153 vs 49,152 in the StarCoder paper (one added token).

n_layers

40recorded · source

hidden_size

6144recorded · source

n_heads

48recorded · source

vocab_size

49153recorded · source

context_length

8192recorded · source

n_kv_heads

1recorded · source

note:  config multi_query: true

positional_encoding

learned absoluterecorded · source

note:  Architecture unchanged from starcoder (fine-tune of StarCoder-15B; config: GPTBigCodeForCausalLM, 40 layers, hidden 6144, 48 heads, multi-query). Value from the StarCoder paper sec. 5.2.

training_data

Code Evol-Instruct: about 78k code instructions, Code Alpaca (~20k) evolved with gpt-3.5-turbo; StarCoder-15B fine-tuned on it.recorded · source

note:  Card: 'WizardCoder-15B-v1.0 trained with 78k evolved code instructions'. Paper: Code Alpaca 'evolved' and StarCoder fine-tuned on the resulting data (see the code-evol-instruct record for the paper's wording on gpt-3.5-turbo).

techniques

primary_sources

https://huggingface.co/WizardLMTeam/WizardCoder-15B-V1.0

https://huggingface.co/WizardLMTeam/WizardCoder-15B-V1.0/raw/9c177589dec389eac2c8de51cbc371d45e47984e/config.json

record_history

date:2026-09-24 · change:ingested as candidate from HF (WizardLMTeam/WizardCoder-15B-V1.0@9c177589dec389eac2c8de51cbc371d45e47984e) · by:ingest_hf.py ·

date:2026-09-24 · change:preparer: developer, license (partial), positional encoding (from StarCoder paper), training data, 2 edges; sources: card, arXiv 2306.08568 · by:claude (preparer, Sonnet 5) ·

date:2026-09-24 · change:reviewed and promoted from staging (2 edge(s) accepted) · by:Wilson Pruitt ·

Parents

Weights descend

• fine_tuned_from → StarCoder declared source Card fine-tuning section: 'We fine-tune StarCoder-15B with the following hyperparameters' and the reproduce command `--model_name_or_path "bigcode/starcoder"`; paper: 'we fine-tune StarCoder'.

Training data

• trained_on → Code Evol-Instruct (WizardCoder) declared source Card: 'trained with 78k evolved code instructions'; paper: Code Alpaca evolved with Code Evol-Instruct, then StarCoder fine-tuned. The `code-evol-instruct` record has availability `unknown` (only third-party reproductions found); this edge is to that record, not to a released file.

Children

No edges recorded.

Read in

No station on the reading path has touched this record yet.