Perry Link

perrylink

PerryLink

AI & ML interests

LLM agents, agent tooling and protocols, tool calling and function calling, MCP, plugin architectures for AI dev tools, local LLM inference, quantization, consumer-GPU deployment, evaluation and benchmarking

Recent Activity

new activity about 6 hours ago

Phocinae/Phocinae-Largha-150M-v1:Demo for this model on Spaces

new activity about 9 hours ago

hugging-apps/phocinae-largha-150m:v1.1 numbers refresh: README + app.py (flip 0.0300 -> 2.17%/2.83%; acc 0.797/0.789 -> 0.906/0.848; ECE 0.1313 -> 0.0168/0.0152; 54% -> -55.0%; 144M -> 144.3M)

updated a model about 10 hours ago

Phocinae/Vitulina-400M-v1

View all activity

Organizations

New activity in Phocinae/Phocinae-Largha-150M-v1 about 6 hours ago

Demo for this model on Spaces

#2 opened about 21 hours ago by

multimodalart

New activity in hugging-apps/phocinae-largha-150m about 9 hours ago

v1.1 numbers refresh: README + app.py (flip 0.0300 -> 2.17%/2.83%; acc 0.797/0.789 -> 0.906/0.848; ECE 0.1313 -> 0.0168/0.0152; 54% -> -55.0%; 144M -> 144.3M)

#1 opened about 9 hours ago by

perrylink

New activity in LocalLLaMA/typed-decisions about 15 hours ago

Submission: Phocinae/Phocinae-Largha-150M-v1 (zero-shot), accuracy 0.797 (en) / 0.789 (zh)

🔥 1

#18 opened 1 day ago by

perrylink

Phocinae-Largha-150M-v1 — test scores (generalist) + reproduction

🔥 1

#15 opened 1 day ago by

perrylink

New activity in Phocinae/Phocinae-Largha-150M-v1 1 day ago

What decisions are you routing locally? (use cases + what breaks first)

#1 opened 1 day ago by

perrylink

New activity in convaiinnovations/laya 5 days ago

CPU latency numbers for laya-multilingual — the model card only has T4

#12 opened 16 days ago by

htqmike

New activity in convaiinnovations/laya 13 days ago

choice:11+ temperature (0.1006) saturates confidence: answers return at 1.00, wrong ones included

👍 3

#2 opened 20 days ago by

DrowsyJungle

Laya-multilingual: a 100-scenario decision-making evaluation

🔥 19

#3 opened 19 days ago by

mghafiri

New activity in convaiinnovations/laya 14 days ago

Zero-shot Laya as an agent step guard: agrees when agents say "done", never escalates destructive commands, blind past 320–768 tokens

👀 2

#15 opened 16 days ago by

sinhal

Perry Link

AI & ML interests

Recent Activity

Organizations

perrylink's activity

Demo for this model on Spaces

v1.1 numbers refresh: README + app.py (flip 0.0300 -> 2.17%/2.83%; acc 0.797/0.789 -> 0.906/0.848; ECE 0.1313 -> 0.0168/0.0152; 54% -> -55.0%; 144M -> 144.3M)

Submission: Phocinae/Phocinae-Largha-150M-v1 (zero-shot), accuracy 0.797 (en) / 0.789 (zh)

Phocinae-Largha-150M-v1 — test scores (generalist) + reproduction

What decisions are you routing locally? (use cases + what breaks first)

CPU latency numbers for laya-multilingual — the model card only has T4

choice:11+ temperature (0.1006) saturates confidence: answers return at 1.00, wrong ones included

Laya-multilingual: a 100-scenario decision-making evaluation

Zero-shot Laya as an agent step guard: agrees when agents say "done", never escalates destructive commands, blind past 320–768 tokens