google/paligemma2-28b-pt-896
Image-Text-to-Text
ā¢
28B
ā¢
Updated
ā¢
237
ā¢
50
Gunnar PRO
grhone
AI & ML interests
None yet
Recent Activity
liked
a model
28 days ago
google/paligemma2-28b-pt-896
upvoted
a
paper
about 1 month ago
VCode: a Multimodal Coding Benchmark with SVG as Symbolic Visual
Representation
upvoted
a
paper
about 1 month ago
Kimi Linear: An Expressive, Efficient Attention Architecture
Organizations
liked
a
model
28 days ago
upvoted
2
papers
about 1 month ago
liked
a
Space
about 2 months ago
liked
a
model
about 2 months ago
fixing hardcoded cuda() for cpu inference
ā¤ļø
5
4
#21 opened about 2 months ago
by
alexgambashidze
upvoted
2
papers
3 months ago
updated
a
model
3 months ago
posted
an
update
3 months ago
Post
241
š Gemma 3 ATSPM Finetune is Live!
I'm excited to share a new model I've been working on, alongside a brand new synthetic dataset. This project targets a very specific, practical problem in transportation.
š¦ Model: grhone/gemma-3-27b-atspm-gguf is a GGUF finetune of Gemma 3 27B Instructions Tuned, specialized for analyzing Automated Traffic Signal Performance Measures (ATSPM) charts.
š¦ Performance: It achieved a 43.97% benchmark score on our custom ATSPM evaluation set, a slight lead over Gemini 2.5 Pro (43.94%) and a significant leap over the base Gemma 3 model (24.76%).
š¦ Training Details: This was achieved by training on approximately 6% of the nodes for 2 epochs using the QLORA method.
š¦ Dataset: grhone/atspm-dataset is a synthetic dataset containing 25,851 QA pairs and nearly 8.5 million tokens, created specifically for this project.
Both the model and the dataset are available for the community to use:
š¤ Model: grhone/gemma-3-27b-atspm-gguf
š Dataset: grhone/atspm-dataset
I'm excited to share a new model I've been working on, alongside a brand new synthetic dataset. This project targets a very specific, practical problem in transportation.
š¦ Model: grhone/gemma-3-27b-atspm-gguf is a GGUF finetune of Gemma 3 27B Instructions Tuned, specialized for analyzing Automated Traffic Signal Performance Measures (ATSPM) charts.
š¦ Performance: It achieved a 43.97% benchmark score on our custom ATSPM evaluation set, a slight lead over Gemini 2.5 Pro (43.94%) and a significant leap over the base Gemma 3 model (24.76%).
š¦ Training Details: This was achieved by training on approximately 6% of the nodes for 2 epochs using the QLORA method.
š¦ Dataset: grhone/atspm-dataset is a synthetic dataset containing 25,851 QA pairs and nearly 8.5 million tokens, created specifically for this project.
Both the model and the dataset are available for the community to use:
š¤ Model: grhone/gemma-3-27b-atspm-gguf
š Dataset: grhone/atspm-dataset
updated
a
dataset
3 months ago
published
a
model
3 months ago
published
a
dataset
3 months ago
liked
2
models
4 months ago
upvoted
2
papers
4 months ago