Contrastive-LM/CLM-v0.1-8B
Text Ranking ā¢ Updated ā¢ 1.27k ā¢ 499
Fontlab Ltd PRO
fontlab
AI & ML interests
Fonts, vector graphics, multilingual text processing
Recent Activity
liked a model about 10 hours ago
Contrastive-LM/CLM-v0.1-8B liked a model 1 day ago
Alissonerdx/BFS-Best-Face-Swap liked a model 3 days ago
coreai-community/GLiNER2-PII-CoreAIOrganizations
liked a
model about 10 hours ago
liked a
model 1 day ago
liked 3
models 3 days ago
reacted to OppaAI's post with š 6 days ago
Post
3148
Benchmark test: Jev vs. Laya-ONNX (multilingual) vs. Harrier OSS 270M embedder š¬
My AI wAIfu (Jetson Orin Nano 8GB) uses Harrier OSS 270M for semantic routing in 2 places. It reads vectors of router prompts (English only) and calculates cosine similarity:
- Quaternary routing: greeting, local chat (no websearch), web chat (needs websearch), or agentic chat
- Agentic routing: which tools in my AI's capability list to use
Benchmarked the 2 most hyped decision models ā Jev and Laya (ONNX, multilingual) ā against Harrier OSS 270M.
Setup: 221 quaternary + 58 capability-trigger examples, leave-one-out eval, argmax, no thresholds.
Results:
ā Harrier-270M (local, cosine): 94.6% / 93.1% accuracy, 17ms P50 ā”
ā Jev API (hosted): 82.4% / 94.8% accuracy, ~195ms P50
ā Laya-ONNX multilingual (fp16, local): 48.0% / 20.7% accuracy, 25-40ms P50
Conclusion:
š« Laya is out of the question. 4 of 7 capability categories at 0.0% accuracy while reporting 80-90% confidence means it needs real training before it's practical.
āļø Jev is a cloud API, not sure it can be trained further. Accuracy is high but not improvable on my end. Latency is ~10x my local embedder (network latency). Input token cost, though small, is still more than $0. Not fully sure about privacy implications either.
ā Embedding is only semantic cosine similarity, not real reasoning. But it's already doing double duty for memory extraction and RAG ā no extra RAM or token cost. Latency is 17ms, accuracy in the 90s%. Even tried Japanese/Chinese prompts, still got high accuracy with only English exemplars.
Bigger advantage: I just add exemplars to boost accuracy. When I add/modify/remove tools ā often ā no retraining needed, just update exemplars, vectors recompute once.
Turns out my self-invented routing method, built ~6 months ago, already solved what these now hyped up models ā beating Jev and Laya on latency and convenience, matching/beating on accuracy. šÆ
My AI wAIfu (Jetson Orin Nano 8GB) uses Harrier OSS 270M for semantic routing in 2 places. It reads vectors of router prompts (English only) and calculates cosine similarity:
- Quaternary routing: greeting, local chat (no websearch), web chat (needs websearch), or agentic chat
- Agentic routing: which tools in my AI's capability list to use
Benchmarked the 2 most hyped decision models ā Jev and Laya (ONNX, multilingual) ā against Harrier OSS 270M.
Setup: 221 quaternary + 58 capability-trigger examples, leave-one-out eval, argmax, no thresholds.
Results:
ā Harrier-270M (local, cosine): 94.6% / 93.1% accuracy, 17ms P50 ā”
ā Jev API (hosted): 82.4% / 94.8% accuracy, ~195ms P50
ā Laya-ONNX multilingual (fp16, local): 48.0% / 20.7% accuracy, 25-40ms P50
Conclusion:
š« Laya is out of the question. 4 of 7 capability categories at 0.0% accuracy while reporting 80-90% confidence means it needs real training before it's practical.
āļø Jev is a cloud API, not sure it can be trained further. Accuracy is high but not improvable on my end. Latency is ~10x my local embedder (network latency). Input token cost, though small, is still more than $0. Not fully sure about privacy implications either.
ā Embedding is only semantic cosine similarity, not real reasoning. But it's already doing double duty for memory extraction and RAG ā no extra RAM or token cost. Latency is 17ms, accuracy in the 90s%. Even tried Japanese/Chinese prompts, still got high accuracy with only English exemplars.
Bigger advantage: I just add exemplars to boost accuracy. When I add/modify/remove tools ā often ā no retraining needed, just update exemplars, vectors recompute once.
Turns out my self-invented routing method, built ~6 months ago, already solved what these now hyped up models ā beating Jev and Laya on latency and convenience, matching/beating on accuracy. šÆ
- 3 replies
liked a
model 6 days ago
liked 3
models 8 days ago
liked 3
Spaces 8 days ago
liked a
model 9 days ago
liked 6
Spaces 13 days ago
LFM2.5 1.2B Thinking WebGPU
š§
130
Run LFM2.5-1.2B-Thinking directly in your browser on WebGPU
GPT OSS WebGPU
ā”
75
Run GPT-OSS-20B locally in your browser on WebGPU
SAM3 Tracker WebGPU
šÆ
32
Segment images with click points and download cutouts
TranslateGemma 4B WebGPU
š
53
Private browser-based translator with 56 languages
LFM2.5-VL-450M WebGPU
š¹
62
Live video captioning and object tracking in your browser
Ternary Bonsai WebGPU
š³
162
Run 1.58-bit Bonsai LLMs locally in your browser on WebGPU