/ Модель

Характеристики

Контекст
131 072 токена
Параметры
7,5 млрд
Размер весов
16,1 ГБ
Архитектура
llama · dense
Базовая модель
DeepSeek-R1-Distill-Llama-8B
Библиотека
transformers
Опубликована
20 янв 2025

LlamaZero-3.1-8B-Experimental-1208

Генерация текстаТекст → текст11 профессий

Слитая языковая модель Llama примерно на 7,5 млрд параметров для диалогов и генерации текста с контекстом до 131 072 токенов.

24скачиваний за 30 дней0лайковвеса обновлены21 янв 2025

Бенчмарки

Как модель справилась с задачами пользователей ailist и что о ней публикуют открытые лидерборды

Тесты пользователейпока нет

Никто ещё не опубликовал тест этой модели. Соберите бенчмарк из своих задач — результат появится здесь.

  1. Добавьте модель в бенчмарк

    Вместе с другими моделями категории «Текст» — сравнение идёт внутри одной категории.

  2. Опишите свои задачи

    Напишите запросы, при необходимости приложите файлы и эталонные ответы.

  3. Запустите и опубликуйте

    Модели ответят на каждую задачу, судья оценит ответы. Опубликованный результат появится на этой странице.

Публичные оценки6

Из открытых лидербордов. Сравнивайте результаты одного теста при одинаковых условиях — отсутствие оценки не значит, что модель хуже.

  • MMLU-PRO

    MMLU-PRO Raw · выше — лучше

    Результат

    0,3

    Open LLM v2 · архив

    Публичная таблицадругая ревизия

  • IFEval

    IFEval Raw · выше — лучше

    Результат

    0,6051

    Open LLM v2 · архив

    Публичная таблицадругая ревизия

  • MATH Lvl 5

    MATH Lvl 5 Raw · выше — лучше

    Результат

    0,108

    Open LLM v2 · архив

    Публичная таблицадругая ревизия

  • GPQA

    GPQA Raw · выше — лучше

    Результат

    0,2685

    Open LLM v2 · архив

    Публичная таблицадругая ревизия

  • BBH

    BBH Raw · выше — лучше

    Результат

    0,4981

    Open LLM v2 · архив

    Публичная таблицадругая ревизия

  • MUSR

    MUSR Raw · выше — лучше

    Результат

    0,382

    Open LLM v2 · архив

    Публичная таблицадругая ревизия

Что умеет модель

Описание составлено по карточке модели и метаданным Hugging Face

LlamaZero 3.1-8B — языковая модель для генерации текста, созданная sethuiyer путём слияния предварительно обученных моделей с помощью mergekit. Она основана на DeepSeek-R1-Distill-Llama-8B и Llama-3.1-8B-Experimental-1206-Instruct; архитектура — плотная Llama с классом LlamaForCausalLM, размер — около 7,5 млрд параметров. Модель предназначена для диалогового взаимодействия и генерации текста; в README приведены результаты оценки на бенчмарках IFEval и BBH.

Метаданные указывают контекст до 131 072 токенов и поддержку чата. В репозитории используются веса safetensors. Лицензия в предоставленных данных не указана.

Сценарии применения

  • ведение диалога с языковой моделью
  • генерация связного текста по запросу
  • ответы на вопросы в текстовом чате

Ограничения

В предоставленных данных не перечислены конкретные языковые ограничения или ограничения применения. Лицензия не указана.

Для каких профессий подходит11

Разделы каталога профессий, в задачах которых используется эта модель

Все профессии

Также подходит для

Карточка модели

README от автора на Hugging Face, как он его опубликовал

Открыть на Hugging Face

LlamaZero 3.1-8B

This is a merge of pre-trained language models created using mergekit.

question = '''
Framework to Reason:
Restate the most critical elements of the scenario concisely to ensure focus on relevant details.
Group evidence by relevance to means, opportunity, and actions:
Focus on the most plausible hypothesis based on the evidence.
Evaluate Evidence for Each Hypothesis:
Draw a Conclusion
Optional: Address Uncertainty or Gaps:

Tips:
Avoid repeating the same point unnecessarily.
Focus on the strongest evidence for each hypothesis.
Briefly mention any uncertainties but do not overcomplicate the analysis.

Prompt:
You are a detective investigating a robbery at a museum. The robbery occurred last night between 10 PM and 11 PM. Here are the key facts:
Security footage shows no signs of entry through the main door.
The back door was locked, but the key is missing from the security guard's desk.
A museum employee, Alex, was seen near the guard's desk at 9:45 PM.
Another employee, Jordan, reported hearing a loud noise from the exhibit hall at 10:15 PM but did not investigate.
The stolen artifact is a small, lightweight sculpture that could fit in a bag.
The alarm system was disabled from the control room, but the exact time of deactivation is unknown.
Task:
Analyze the situation and determine who is most likely responsible for the robbery. Include your reasoning based on the evidence provided.
'''
response = assistant.generate_response(question,  max_new_tokens=2048)
print(response)
To determine the most likely responsible person for the robbery, I will analyze the evidence provided.

**Critical Elements:**
- The robbery occurred between 10 PM and 11 PM.
- No signs of entry through the main door.
- The back door was locked, but the key is missing from the security guard's desk.
- Alex was seen near the guard's desk at 9:45 PM.
- Jordan reported hearing a loud noise from the exhibit hall at 10:15 PM but did not investigate.
- The stolen artifact is small and lightweight.
- The alarm system was disabled from the control room, but the exact time of deactivation is unknown.

**Grouping Evidence:**
- **Means:** Access to the back door key or knowledge of an alternate entry point.
- **Opportunity:** Being in the vicinity of the back door or the control room during the time of the robbery.
- **Actions:** Disabling the alarm system, entering the exhibit hall, and taking the sculpture.

**Most Plausible Hypothesis:**
Based on the evidence, the most likely responsible person is Alex. Here's why:
- Alex was seen near the guard's desk at 9:45 PM, which is around the time the robbery occurred. This suggests Alex had access to the back door key or was involved in disabling the alarm.
- The back door key was missing from the security guard's desk, indicating that someone with access to it was involved. Alex's presence near the desk at 9:45 PM suggests they may have taken the key.
- Jordan heard a noise at 10:15 PM, which is around the time the robbery occurred. However, Jordan did not investigate, which suggests they may not have been the one to enter or they may have been distracted.
- The alarm system was disabled, which could have been done by Alex or someone else. However, since Alex was near the guard's desk, it's likely they disabled the alarm and then entered through the back door.
- The sculpture was stolen, and Alex had the opportunity to take it after disabling the alarm and entering through the back door.

**Evaluation of Evidence:**
The evidence strongly suggests Alex was involved. Alex's presence near the guard's desk at 9:45 PM and the missing key indicate they had the means to enter the back door. The timing of their presence is also consistent with the time of the robbery. Additionally, Alex would have had the opportunity to take the sculpture and disable the alarm.

**Conclusion:**
Based on the evidence, Alex is the most likely responsible for the robbery.

**Uncertainty or Gaps:**
- The exact time the alarm was disabled is unknown, but since Alex was seen near the desk at 9:45 PM, it's likely they disabled the alarm shortly before the robbery. This is a minor uncertainty but does not affect the conclusion.
- Jordan's lack of investigation could be due to being asleep or distracted, but this does not necessarily rule out Alex as the perpetrator.
- There is no direct evidence linking Alex to the actual theft, but the timing and access to the key make it highly probable.

Therefore, the most likely responsible person is Alex.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

MetricValue
Avg.21.77
IFEval (0-Shot)60.51
BBH (3-Shot)28.61
MATH Lvl 5 (4-Shot)9.67
GPQA (0-shot)2.46
MuSR (0-shot)7.15
MMLU-PRO (5-shot)22.22

Как запустить

На своей GPU в ailist. Цена — по текущим предложениям провайдеров

GPU в ailist

Универсальный рантайм · transformers

Можно запустить

Подбираем GPU…

Класс архитектуры
LlamaForCausalLM
Ревизия весов
bf7a0ab

0 моделей