Gryphe/Sonnet3.5-SlimOrcaDedupCleaned
Viewer
•
Updated
•
181k
•
222
•
85
Kiran Kamble
kiranr
AI & ML interests
nlp,llm
Recent Activity
liked
a dataset
16 days ago
Gryphe/Sonnet3.5-SlimOrcaDedupCleaned
liked
a dataset
16 days ago
Magpie-Align/Magpie-Llama-3.1-Pro-MT-300K-Filtered
liked
a dataset
16 days ago
virattt/financial-qa-10K
Organizations
kiranr's activity
liked
4
datasets
16 days ago
updated
a
model
26 days ago
reacted to
melisa's
post with 🔥
3 months ago
Post
2981
🔥 Introducing "Writing in the Margins (WiM)" - better inference pattern for long context LLMs that solves the Lost-in-the-Middle problem 🔥
Paper page: Writing in the Margins: Better Inference Pattern for Long Context Retrieval (2408.14906)
TL;DR
Make your model write "margin notes" as you chunk prefill the KV cache. Then ask it reread all notes before it speaks up.
Works with humans, works with AI 🤖
WiM leverages the chunked prefill of the key-value cache, which concurrently generates query-based extractive summaries at each step of the prefill that are subsequently reintegrated at the end of the computation. We term these intermediate outputs “margins”, drawing inspiration from the practice of making margin notes for improved comprehension of long contexts in human reading. We show that this technique, which adds only minimal additional computation, significantly improves LLMs long context reasoning capabilities.
Think: Every chunk has a chance to be attended to/ be at the end of the context at least once. 🎉
📊 Results:
- An average accuracy boost of 7.5% in multi-hop reasoning tasks like HotpotQA and MultiHop-RAG.
- Even a 30% increase in F1-score for summarisation-like tasks (CWE).
Plus, WiM fits seamlessly into interactive applications (think: progress bar!). It can provide real-time progress updates during data retrieval and integration, making it user-friendly and transparent - a stark contrast to feeding 1mln tokens to an LLMs and waiting 6 min for the first token. 🤯
👩💻🧑💻 Check it out and contribute to our open-source project here: https://github.com/writer/writing-in-the-margins
🧠 More about chunked prefill: https://docs.vllm.ai/en/latest/models/performance.html#chunked-prefill
Paper page: Writing in the Margins: Better Inference Pattern for Long Context Retrieval (2408.14906)
TL;DR
Make your model write "margin notes" as you chunk prefill the KV cache. Then ask it reread all notes before it speaks up.
Works with humans, works with AI 🤖
WiM leverages the chunked prefill of the key-value cache, which concurrently generates query-based extractive summaries at each step of the prefill that are subsequently reintegrated at the end of the computation. We term these intermediate outputs “margins”, drawing inspiration from the practice of making margin notes for improved comprehension of long contexts in human reading. We show that this technique, which adds only minimal additional computation, significantly improves LLMs long context reasoning capabilities.
Think: Every chunk has a chance to be attended to/ be at the end of the context at least once. 🎉
📊 Results:
- An average accuracy boost of 7.5% in multi-hop reasoning tasks like HotpotQA and MultiHop-RAG.
- Even a 30% increase in F1-score for summarisation-like tasks (CWE).
Plus, WiM fits seamlessly into interactive applications (think: progress bar!). It can provide real-time progress updates during data retrieval and integration, making it user-friendly and transparent - a stark contrast to feeding 1mln tokens to an LLMs and waiting 6 min for the first token. 🤯
👩💻🧑💻 Check it out and contribute to our open-source project here: https://github.com/writer/writing-in-the-margins
🧠 More about chunked prefill: https://docs.vllm.ai/en/latest/models/performance.html#chunked-prefill
- 2 replies
updated
a
collection
3 months ago
upvoted
a
paper
3 months ago
authored
a
paper
3 months ago
commented
a
paper
3 months ago
upvoted
an
article
3 months ago
Article
Using Writer Framework with Hugging Face Spaces
By
•
•
30
updated
a
collection
4 months ago
upvoted
2
papers
4 months ago
updated
a
collection
4 months ago
upvoted
a
paper
4 months ago
Access the model via Ollama
1
#3 opened 4 months ago
by
gileneo
gileneo
reacted to
samjulien's
post with 🔥
4 months ago
Post
1897
🔥 Today, Writer dropped Palmyra-Med-70b and Palmyra-Fin-70b, two new domain-specific models that are setting a new standard for medical and financial model performance.
TL;DR
Palmyra-Med-70b
🔢 8k and 32k versions available
🚀 MMLU performance of ~86%, outperforming other top models
👨⚕️ Great for diagnosing, planning treatments, medical research, insurance coding and billing
📃 Open-model license for non-commercial use cases
🤗 Available on Hugging Face: Writer/Palmyra-Med-70B
💾 Live on NVIDIA NIM: https://build.nvidia.com/writer/palmyra-med-70b
Palmyra-Fin-70b
🚀 Passed the CFA Level III exam with a 73% score — the first model to do so
💸 Skilled at complex tasks like investment research, financial analysis, and sentiment analysis
📈 Outperformed other top models on a long-fin-eval test of real-world use cases
📃 Open-model license for non-commercial use cases
🤗 Available on Hugging Face: Writer/Palmyra-Fin-70B-32K
💾 Live on NVIDIA NIM: https://build.nvidia.com/writer/palmyra-fin-70b-32k
Try them out and let us know what you think!
TL;DR
Palmyra-Med-70b
🔢 8k and 32k versions available
🚀 MMLU performance of ~86%, outperforming other top models
👨⚕️ Great for diagnosing, planning treatments, medical research, insurance coding and billing
📃 Open-model license for non-commercial use cases
🤗 Available on Hugging Face: Writer/Palmyra-Med-70B
💾 Live on NVIDIA NIM: https://build.nvidia.com/writer/palmyra-med-70b
Palmyra-Fin-70b
🚀 Passed the CFA Level III exam with a 73% score — the first model to do so
💸 Skilled at complex tasks like investment research, financial analysis, and sentiment analysis
📈 Outperformed other top models on a long-fin-eval test of real-world use cases
📃 Open-model license for non-commercial use cases
🤗 Available on Hugging Face: Writer/Palmyra-Fin-70B-32K
💾 Live on NVIDIA NIM: https://build.nvidia.com/writer/palmyra-fin-70b-32k
Try them out and let us know what you think!
-
- 2 replies
Dataset sources
2
#1 opened 4 months ago
by
MatVet
MatVet