Llama 3.1 8B Instruct
An 8-billion-parameter instruction-tuned large language model released by Meta as part of the Llama 3.1 series, supporting multilingual use and long context.
Timeline (last 90 days)
On the AlpacaEval 2.0 benchmark (with GPT-4o as the judge), after three rounds of DNO-Prct iteration, the length-controlled (LC) win rate of Llama 3.1 8B Instruct against GPT-4o rose from 31.7% to 45.3%.
The tuned Llama 3.1 8B Instruct surpasses Llama 3.1 70B Instruct on all reported win-rate metrics.
The research team independently reimplemented the DNO-Prct pipeline rather than directly using the original authors' code, and validated it on Llama 3.1 8B Instruct and Mistral 7B Instruct v0.1.
During writing, a local WebSocket test loop was actually run; the bridge completed the config handshake, spoke the first sentence, and streamed back a coherent 18 chunks from llama-3.1-8b-instruct, and ping and interruption handling both worked as documented.
Connecting an open-source model to the Retell Custom LLM slot requires only a bridge service of about 90 lines.
When using llama-3.1-8b-instruct bridged to EcoHash, calculated at about 2000 input tokens and 250 output tokens per minute for a typical customer service call at a unit price of $0.10/1M, the token cost is about $0.0003/minute.
The bridge is located in the integrations/retell directory of ecohash-examples, uses the llama-3.1-8b-instruct model by default, and is priced at $0.10 / 1M tokens for both input and output.
The measured performance of llama-3.1-8b-instruct on EcoHash is a TTFT of 205 ms (p95) and about 143 tokens/second for a single stream.
All Facts (5)
On the AlpacaEval 2.0 benchmark (with GPT-4o as the judge), after three rounds of DNO-Prct iteration, the length-controlled (LC) win rate of Llama 3.1 8B Instruct against GPT-4o rose from 31.7% to 45.3%.
50%UnverifiedThe tuned Llama 3.1 8B Instruct surpasses Llama 3.1 70B Instruct on all reported win-rate metrics.
50%UnverifiedConnecting an open-source model to the Retell Custom LLM slot requires only a bridge service of about 90 lines.
50%UnverifiedThe bridge is located in the integrations/retell directory of ecohash-examples, uses the llama-3.1-8b-instruct model by default, and is priced at $0.10 / 1M tokens for both input and output.
50%UnverifiedThe research team independently reimplemented the DNO-Prct pipeline rather than directly using the original authors' code, and validated it on Llama 3.1 8B Instruct and Mistral 7B Instruct v0.1.
50%