Skip to page content
SAVRN
Search Plan a project

Open-weight model

gandalf-qwen2.5-3b-lora

by Henry Tan henryatharvard/gandalf-qwen2.5-3b-lora

gandalf-qwen2.5-3b-lora is an open-weight model from Henry Tan. Its published files total 6.3 GB.

A post-training study: teach a 3B model to talk like Gandalf (voice derived from the film transcripts) without losing GSM8K accuracy. Every adapter here is LoRA r=32 on all seven projections of every layer, bf16.

Parameters—
Context—
Weights6.0 GB
License—
AccessOpen weights
Monthly Downloads—

Model Card

A post-training study: teach a 3B model to talk like Gandalf (voice derived from the film transcripts) without losing GSM8K accuracy. Every adapter here is LoRA r=32 on all seven projections of every layer, bf16. All were evaluated on the same 200 GSM8K test questions (rule-based verifier, greedy) and 60 held-out open prompts (voice 1 to 5 from an LLM judge). The prompted base model scores 0.855 accuracy, 1.0 voice. Best checkpoint: 010/tower-dpo, the voice tower after DPO (restatement accuracy 0.870, faithful 0.77, voice on math 4.36). Best single model: 004/D320-rlvr (accuracy 0.885, voice 3.67). System prompt used in training and evaluation: the Gandalf persona prompt built from…

Excerpt from the card by Henry Tan.

Identity and Version

Repository
henryatharvard/gandalf-qwen2.5-3b-lora
Publisher
Henry Tan
Task
Not stated by the source
Modality
Other
Library
peft
Parameters
Not stated by the source
Languages
Not stated by the source
Revision
5c8aa7f0db4c68987447d63b3cd5693367ebd8d8
First published
2026-09-11
Last updated
2026-09-20

Files and Weights

127 files, 6.3 GB in total. The weights are 25 files totalling 6.0 GB in safetensors.

Weights25 files · 6.0 GB
Configuration25 files · 28.9 KB
Tokenizer50 files · 285.6 MB
Documentation1 file · 5.2 KB
Other25 files · 62.7 KB
Repository1 file · 3.0 KB
Every file
FileTypeSizeSHA-256
003/combined/adapter_model.safetensorsWeights239.5 MB 085ab6ae351a
003/rlaif/adapter_model.safetensorsWeights239.5 MB 9f65993334a1
003/rlaif_naked/adapter_model.safetensorsWeights239.5 MB 132622174592
003/rlvr/adapter_model.safetensorsWeights239.5 MB 1d7252b9f4a7
003/sft/adapter_model.safetensorsWeights239.5 MB b1572b8bef4b
004/A/adapter_model.safetensorsWeights239.5 MB 514738fddfa2
004/B/adapter_model.safetensorsWeights239.5 MB 1cd6d6783763
004/C/adapter_model.safetensorsWeights239.5 MB 8c3493b9c1cf
004/D160/adapter_model.safetensorsWeights239.5 MB 2c89f9baa13f
004/D320-combined/adapter_model.safetensorsWeights239.5 MB e7f31e1363ec
004/D320-rlaif/adapter_model.safetensorsWeights239.5 MB 97bf8de4fe6e
004/D320-rlvr/adapter_model.safetensorsWeights239.5 MB e017f3e6c810
004/D320/adapter_model.safetensorsWeights239.5 MB 0fc930b86cf3
004/D511/adapter_model.safetensorsWeights239.5 MB 4698a5dd1fd0
008/rlaif/adapter_model.safetensorsWeights239.5 MB 24bbe586729d
008/rlvr/adapter_model.safetensorsWeights239.5 MB 7368fb92daa9
008/sft/adapter_model.safetensorsWeights239.5 MB 03c8725c111a
010/D320-dpo/adapter_model.safetensorsWeights239.5 MB 25cda62c90dc
010/D320-kto/adapter_model.safetensorsWeights239.5 MB cf8ea410163f
010/tower-dpo/adapter_model.safetensorsWeights239.5 MB 3c9658861d47
combined/adapter_model.safetensorsWeights239.5 MB af245c2436ce
rlaif/adapter_model.safetensorsWeights239.5 MB 0e2a527b3c6b
rlaif_naked/adapter_model.safetensorsWeights239.5 MB 3e9d1e6d49fe
rlvr/adapter_model.safetensorsWeights239.5 MB 62330ae6e54a
sft/adapter_model.safetensorsWeights239.5 MB 9542f511efab
003/combined/adapter_config.jsonConfiguration1.2 KB —
003/rlaif/adapter_config.jsonConfiguration1.2 KB —
003/rlaif_naked/adapter_config.jsonConfiguration1.2 KB —
003/rlvr/adapter_config.jsonConfiguration1.2 KB —
003/sft/adapter_config.jsonConfiguration1.2 KB —
004/A/adapter_config.jsonConfiguration1.2 KB —
004/B/adapter_config.jsonConfiguration1.2 KB —
004/C/adapter_config.jsonConfiguration1.2 KB —
004/D160/adapter_config.jsonConfiguration1.2 KB —
004/D320-combined/adapter_config.jsonConfiguration1.2 KB —
004/D320-rlaif/adapter_config.jsonConfiguration1.2 KB —
004/D320-rlvr/adapter_config.jsonConfiguration1.2 KB —
004/D320/adapter_config.jsonConfiguration1.2 KB —
004/D511/adapter_config.jsonConfiguration1.2 KB —
008/rlaif/adapter_config.jsonConfiguration1.2 KB —
008/rlvr/adapter_config.jsonConfiguration1.2 KB —
008/sft/adapter_config.jsonConfiguration1.2 KB —
010/D320-dpo/adapter_config.jsonConfiguration1.2 KB —
010/D320-kto/adapter_config.jsonConfiguration1.2 KB —
010/tower-dpo/adapter_config.jsonConfiguration1.2 KB —
combined/adapter_config.jsonConfiguration1.2 KB —
rlaif/adapter_config.jsonConfiguration1.2 KB —
rlaif_naked/adapter_config.jsonConfiguration1.2 KB —
rlvr/adapter_config.jsonConfiguration1.2 KB —
sft/adapter_config.jsonConfiguration1.2 KB —
README.mdDocumentation5.2 KB —
003/combined/chat_template.jinjaOther2.5 KB —
003/rlaif/chat_template.jinjaOther2.5 KB —
003/rlaif_naked/chat_template.jinjaOther2.5 KB —
003/rlvr/chat_template.jinjaOther2.5 KB —
003/sft/chat_template.jinjaOther2.5 KB —
004/A/chat_template.jinjaOther2.5 KB —
004/B/chat_template.jinjaOther2.5 KB —
004/C/chat_template.jinjaOther2.5 KB —
004/D160/chat_template.jinjaOther2.5 KB —
004/D320-combined/chat_template.jinjaOther2.5 KB —
004/D320-rlaif/chat_template.jinjaOther2.5 KB —
004/D320-rlvr/chat_template.jinjaOther2.5 KB —
004/D320/chat_template.jinjaOther2.5 KB —
004/D511/chat_template.jinjaOther2.5 KB —
008/rlaif/chat_template.jinjaOther2.5 KB —
008/rlvr/chat_template.jinjaOther2.5 KB —
008/sft/chat_template.jinjaOther2.5 KB —
010/D320-dpo/chat_template.jinjaOther2.5 KB —
010/D320-kto/chat_template.jinjaOther2.5 KB —
010/tower-dpo/chat_template.jinjaOther2.5 KB —
combined/chat_template.jinjaOther2.5 KB —
rlaif/chat_template.jinjaOther2.5 KB —
rlaif_naked/chat_template.jinjaOther2.5 KB —
rlvr/chat_template.jinjaOther2.5 KB —
sft/chat_template.jinjaOther2.5 KB —
.gitattributesRepository3.0 KB —
003/combined/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
003/combined/tokenizer_config.jsonTokenizer694 B —
003/rlaif/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
003/rlaif/tokenizer_config.jsonTokenizer694 B —
003/rlaif_naked/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
003/rlaif_naked/tokenizer_config.jsonTokenizer694 B —
003/rlvr/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
003/rlvr/tokenizer_config.jsonTokenizer694 B —
003/sft/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
003/sft/tokenizer_config.jsonTokenizer694 B —
004/A/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/A/tokenizer_config.jsonTokenizer694 B —
004/B/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/B/tokenizer_config.jsonTokenizer694 B —
004/C/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/C/tokenizer_config.jsonTokenizer694 B —
004/D160/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/D160/tokenizer_config.jsonTokenizer694 B —
004/D320-combined/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/D320-combined/tokenizer_config.jsonTokenizer694 B —
004/D320-rlaif/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/D320-rlaif/tokenizer_config.jsonTokenizer694 B —
004/D320-rlvr/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/D320-rlvr/tokenizer_config.jsonTokenizer694 B —
004/D320/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/D320/tokenizer_config.jsonTokenizer694 B —
004/D511/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
004/D511/tokenizer_config.jsonTokenizer694 B —
008/rlaif/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
008/rlaif/tokenizer_config.jsonTokenizer694 B —
008/rlvr/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
008/rlvr/tokenizer_config.jsonTokenizer694 B —
008/sft/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
008/sft/tokenizer_config.jsonTokenizer694 B —
010/D320-dpo/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
010/D320-dpo/tokenizer_config.jsonTokenizer694 B —
010/D320-kto/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
010/D320-kto/tokenizer_config.jsonTokenizer694 B —
010/tower-dpo/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
010/tower-dpo/tokenizer_config.jsonTokenizer694 B —
combined/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
combined/tokenizer_config.jsonTokenizer694 B —
rlaif/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
rlaif/tokenizer_config.jsonTokenizer694 B —
rlaif_naked/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
rlaif_naked/tokenizer_config.jsonTokenizer694 B —
rlvr/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
rlvr/tokenizer_config.jsonTokenizer694 B —
sft/tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
sft/tokenizer_config.jsonTokenizer694 B —

License and Download

License
Not stated by the source
Access
Open weights, no gate
Download size
6.0 GB
Download from Henry Tan

Released by Henry Tan through its official repository on Hugging Face.

Built From

Memory Requirements

PrecisionWeights in memory
As published6.0 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.