woot woot! First finetune of GliNER-Decide!
September 25, 2026 at 1:16 PM
woot woot! First finetune of GliNER-Decide!
This tweet appeared under this Techmeme headline:
Brandon Carl / @brandonjcarl:
More Jev competitors out. I've added GliNER 2.5 Decision and CLM 8B to my tests. Jev is handsomely outperforming everything. [image] [embedded post]
Brandon Carl / @brandonjcarl:
More Jev competitors out. I've added GliNER 2.5 Decision and CLM 8B to my tests. Jev is handsomely outperforming everything. [image] [embedded post]
September 24, 2026 at 9:17 PM
This tweet appeared under this Techmeme headline:
Brandon Carl / @brandonjcarl:
More Jev competitors out. I've added GliNER 2.5 Decision and CLM 8B to my tests. Jev is handsomely outperforming everything. [image] [embedded post]
Brandon Carl / @brandonjcarl:
More Jev competitors out. I've added GliNER 2.5 Decision and CLM 8B to my tests. Jev is handsomely outperforming everything. [image] [embedded post]
hm maybe the poster mixed up jev & gliner reading the charts? does look less clear-cut than he claimed
also it's very weird nobody's comparing this or jev to normal autoregressive models with structured generation
also it's very weird nobody's comparing this or jev to normal autoregressive models with structured generation
September 20, 2026 at 12:24 AM
hm maybe the poster mixed up jev & gliner reading the charts? does look less clear-cut than he claimed
also it's very weird nobody's comparing this or jev to normal autoregressive models with structured generation
also it's very weird nobody's comparing this or jev to normal autoregressive models with structured generation
Gemini labeled 4,290 Reddit comments for $9, then a local GLiNER model reached 0.83 F1 against those labels. I like the economics, but I would not trust that score without a small human-checked test set.
https://markhuang.ai/news/gemini-ner-teacher-needs-audit
https://markhuang.ai/news/gemini-ner-teacher-needs-audit
September 18, 2026 at 2:45 AM
Gemini labeled 4,290 Reddit comments for $9, then a local GLiNER model reached 0.83 F1 against those labels. I like the economics, but I would not trust that score without a small human-checked test set.
https://markhuang.ai/news/gemini-ner-teacher-needs-audit
https://markhuang.ai/news/gemini-ner-teacher-needs-audit
Jev fauxpen source to GLiNER disenchantment arc speedrun any %
September 17, 2026 at 6:40 PM
Jev fauxpen source to GLiNER disenchantment arc speedrun any %
Destilación de Gemini: cómo reemplazarla por US$9
¿Podés reemplazar la API de Gemini por apenas US$9? La destilación de Gemini de Peter Vijeh llegó a 0,83 F1 tras cinco entrenamientos fallidos.
#destilacióndegemini #gliner #gemini31pro #finetuning #ner
¿Podés reemplazar la API de Gemini por apenas US$9? La destilación de Gemini de Peter Vijeh llegó a 0,83 F1 tras cinco entrenamientos fallidos.
#destilacióndegemini #gliner #gemini31pro #finetuning #ner
Destilación de Gemini: cómo reemplazarla por US$9
Peter Vijeh entrenó un GLiNER local con etiquetas de Gemini 3.1 Pro y llegó a 0,83 F1 por US$11,50, tras cinco corridas de entrenamiento fallidas.
blog.donweb.com
September 17, 2026 at 5:24 PM
[JP] GeminiでRedditコメントのラベル付けを$9で成功させた話
[EN] Successfully Labeling Reddit Comments with Gemini for Just $9
https://ai-minor.com/blog/en/2026-09-17-1789654706650-i_had_gemini_train_its_own_replacement_for__9
#Gemini #GLiNER #NER #AI #Tech
[EN] Successfully Labeling Reddit Comments with Gemini for Just $9
https://ai-minor.com/blog/en/2026-09-17-1789654706650-i_had_gemini_train_its_own_replacement_for__9
#Gemini #GLiNER #NER #AI #Tech
Successfully Labeling Reddit Comments with Gemini for Just $9
Successfully Labeling Reddit Comments with Gemini for Just $9
ai-minor.com
September 17, 2026 at 4:45 PM
I had Gemini train its own replacement for $9
I had Gemini train its own replacement for $9
Gemini 3.1 Pro labeled 4,290 Reddit comments for $9; a fine-tuned GLiNER model now tags brands, models and materials locally at 0.83 F1 against those labels. Five of ten training runs produced nothing because of a tensor I misread as an attention mask; the loss went flat and no error named the tenso
www.petervijeh.com
September 17, 2026 at 3:40 PM
I had Gemini train its own replacement for $9
I had Gemini train its own replacement for $9 https://www.petervijeh.com/projects/reddit-ner (https://news.ycombinator.com/item?id=49740330)
I had Gemini train its own replacement for $9
Gemini 3.1 Pro labeled 4,290 Reddit comments for $9; a fine-tuned GLiNER model now tags brands, models and materials locally at 0.83 F1 against those labels. Five of ten training runs produced nothing because of a tensor I misread as an attention mask; the loss went flat and no error named the tensor.
www.petervijeh.com
September 17, 2026 at 3:03 PM
I had Gemini train its own replacement for $9 https://www.petervijeh.com/projects/reddit-ner (https://news.ycombinator.com/item?id=49740330)
I had Gemini train its own replacement for $9 | Discussion
I had Gemini train its own replacement for $9
Gemini 3.1 Pro labeled 4,290 Reddit comments for $9; a fine-tuned GLiNER model now tags brands, models and materials locally at 0.83 F1 against those labels. Five of ten training runs produced nothing because of a tensor I misread as an attention mask; the loss went flat and no error named the tensor.
www.petervijeh.com
September 17, 2026 at 2:40 PM
I had Gemini train its own replacement for $9 | Discussion
I had Gemini train its own replacement for $9 https://www.petervijeh.com/projects/reddit-ner (https://news.ycombinator.com/item?id=49740330)
I had Gemini train its own replacement for $9
Gemini 3.1 Pro labeled 4,290 Reddit comments for $9; a fine-tuned GLiNER model now tags brands, models and materials locally at 0.83 F1 against those labels. Five of ten training runs produced nothing because of a tensor I misread as an attention mask; the loss went flat and no error named the tensor.
www.petervijeh.com
September 17, 2026 at 2:23 PM
I had Gemini train its own replacement for $9 https://www.petervijeh.com/projects/reddit-ner (https://news.ycombinator.com/item?id=49740330)
Fastino just launched GLiNER 2.5, swapping clunky span enumeration for sleek boundary prediction. Faster, cleaner entity‑relation decoding on any encoder model. Curious how this upgrades your NER pipeline? Dive in! #GLiNER2_5 #BoundaryPrediction #InfoExtraction
🔗 aidailypost.com/news/fastino...
🔗 aidailypost.com/news/fastino...
August 25, 2026 at 12:09 AM
Fastino just launched GLiNER 2.5, swapping clunky span enumeration for sleek boundary prediction. Faster, cleaner entity‑relation decoding on any encoder model. Curious how this upgrades your NER pipeline? Dive in! #GLiNER2_5 #BoundaryPrediction #InfoExtraction
🔗 aidailypost.com/news/fastino...
🔗 aidailypost.com/news/fastino...
44M params, Apache 2.0, clinical PII token-classification — a light on-ramp to Huawei’s Ascend via Modelers.cn.
Thin card: no benchmarks, no data notes, no NPU proof. Messy or multilingual records could break it fast.
Hard to see this beating a tuned spaCy or GLiNER pipeline for production...
Thin card: no benchmarks, no data notes, no NPU proof. Messy or multilingual records could break it fast.
Hard to see this beating a tuned spaCy or GLiNER pipeline for production...
Spotting Patient Data in Clinical Text: OpenMed's Compact PII Detector Lands on Modelers.cn
aichina.news
August 13, 2026 at 1:39 PM
44M params, Apache 2.0, clinical PII token-classification — a light on-ramp to Huawei’s Ascend via Modelers.cn.
Thin card: no benchmarks, no data notes, no NPU proof. Messy or multilingual records could break it fast.
Hard to see this beating a tuned spaCy or GLiNER pipeline for production...
Thin card: no benchmarks, no data notes, no NPU proof. Messy or multilingual records could break it fast.
Hard to see this beating a tuned spaCy or GLiNER pipeline for production...
Apache-2.0, 125M params, MLX-ready for Apple Silicon — a rare low-resource play for Telugu medical PII redaction on Huawei’s Modelers.cn.
Yet zero benchmark scores or data-card detail means you’re benchmarking it yourself.
Hard to see this beating GLiNER or a fine-tuned Qwen in production...
Yet zero benchmark scores or data-card detail means you’re benchmarking it yourself.
Hard to see this beating GLiNER or a fine-tuned Qwen in production...
Telugu Medical PII: A 125M Model for Privacy on Macs
aichina.news
August 12, 2026 at 6:55 PM
Apache-2.0, 125M params, MLX-ready for Apple Silicon — a rare low-resource play for Telugu medical PII redaction on Huawei’s Modelers.cn.
Yet zero benchmark scores or data-card detail means you’re benchmarking it yourself.
Hard to see this beating GLiNER or a fine-tuned Qwen in production...
Yet zero benchmark scores or data-card detail means you’re benchmarking it yourself.
Hard to see this beating GLiNER or a fine-tuned Qwen in production...
Apache-2.0 66M Portuguese clinical PII model: small enough for Ascend NPUs, permissive for commercial use. But zero benchmarks, undisclosed data—accuracy unproven. Hard to see this displacing spaCy or GLiNER until precision/recall lands on real notes.
A Lightweight Portuguese PII Detector from the Ascend Ecosystem: OpenMed-66M
aichina.news
August 11, 2026 at 12:15 PM
Apache-2.0 66M Portuguese clinical PII model: small enough for Ascend NPUs, permissive for commercial use. But zero benchmarks, undisclosed data—accuracy unproven. Hard to see this displacing spaCy or GLiNER until precision/recall lands on real notes.
I do think it's possible he gets more conservative in a kind of 'Rowling (or more likely Gliner)' path towards being against the 'radical left' but it's always going to be based on his pro-Israel jingoism and general racism over the superiority of 'The West'
July 22, 2026 at 9:25 PM
I do think it's possible he gets more conservative in a kind of 'Rowling (or more likely Gliner)' path towards being against the 'radical left' but it's always going to be based on his pro-Israel jingoism and general racism over the superiority of 'The West'
Open source. Integrates with spaCy, Hugging Face, Ray Serve and ONNX Runtime. Fine-tuning and deployment docs on the listing. www.everydev.ai/tools/gliner
July 1, 2026 at 11:57 PM
Open source. Integrates with spaCy, Hugging Face, Ray Serve and ONNX Runtime. Fine-tuning and deployment docs on the listing. www.everydev.ai/tools/gliner
Labeling training data for every new entity type is a grind. GLiNER does zero-shot NER with bidirectional transformers, so you can extract names, dates, PII or custom entities without task-specific training. 100+ languages supported. More specs in the next post. #MLOps
July 1, 2026 at 11:57 PM
Labeling training data for every new entity type is a grind. GLiNER does zero-shot NER with bidirectional transformers, so you can extract names, dates, PII or custom entities without task-specific training. 100+ languages supported. More specs in the next post. #MLOps
June 14, 2026 at 7:30 AM
NVIDIA shipped `gliner-PII` (gliner_large, ~300M) with weights but no training recipe.
We took `gliner_small-v2.1` (~50M), trained on the full 100k Nemotron-PII set, 3 epochs, 4×A100, bf16.
Result: a statistical tie with a model 3× its size.
We took `gliner_small-v2.1` (~50M), trained on the full 100k Nemotron-PII set, 3 epochs, 4×A100, bf16.
Result: a statistical tie with a model 3× its size.
June 6, 2026 at 9:31 AM
NVIDIA shipped `gliner-PII` (gliner_large, ~300M) with weights but no training recipe.
We took `gliner_small-v2.1` (~50M), trained on the full 100k Nemotron-PII set, 3 epochs, 4×A100, bf16.
Result: a statistical tie with a model 3× its size.
We took `gliner_small-v2.1` (~50M), trained on the full 100k Nemotron-PII set, 3 epochs, 4×A100, bf16.
Result: a statistical tie with a model 3× its size.
GLiNER Guard (GLiGuard): один schema-driven энкодер вместо зоопарка LLM-гардрейлов Деплоите LLM? Значит, обвешиваете её гардами....
#GLiNER #Guard #GLiGuard #GLiNER #2 #guardrails #PII #zero-shot #безопасность #LLM #обработка
Origin | Interest | Match
#GLiNER #Guard #GLiGuard #GLiNER #2 #guardrails #PII #zero-shot #безопасность #LLM #обработка
Origin | Interest | Match
May 19, 2026 at 11:27 PM
GLiNER Guard (GLiGuard): один schema-driven энкодер вместо зоопарка LLM-гардрейлов Деплоите LLM? Значит, обвешиваете её гардами....
#GLiNER #Guard #GLiGuard #GLiNER #2 #guardrails #PII #zero-shot #безопасность #LLM #обработка
Origin | Interest | Match
#GLiNER #Guard #GLiGuard #GLiNER #2 #guardrails #PII #zero-shot #безопасность #LLM #обработка
Origin | Interest | Match
GLiNER Large-v2.1 offers zero-shot NER for arbitrary labels, commercially licensed and optimised for Huawei Ascend. Its 512-token context window may bottleneck complex document analysis. We question if its flexibility truly outweighs domain-specific BERTs for raw speed in production.
One NER Model to Rule Them All: Exploring GLiNER Large-v2.1 on Modelers.cn
aichina.news
May 19, 2026 at 12:05 PM
GLiNER Large-v2.1 offers zero-shot NER for arbitrary labels, commercially licensed and optimised for Huawei Ascend. Its 512-token context window may bottleneck complex document analysis. We question if its flexibility truly outweighs domain-specific BERTs for raw speed in production.
Fastino Labs, Creator of GLiNER, Releases Two State-of-the-Art Language Models 1,000x Smaller Than Frontier
Fastino Labs, Creator of GLiNER, Releases Two State-of-the-Art Language Models 1,000x Smaller Than Frontier - BeyondLink
The two open-source language models are 1000x smaller than large models from companies like OpenAI and Anthropic, with higher accuracy.
beyondlink.live
May 14, 2026 at 1:23 PM
Fastino Labs, Creator of GLiNER, Releases Two State-of-the-Art Language Models 1,000x Smaller Than Frontier
