Skip to content
Language models

SO-CODE L8-CODE

глубина помогает, но с убывающей отдачей

2 min read

Исследовательская модель от dimitarpg13 для генерации кода на Python. Это SO-CODE с восемью слоями и шагом Δt = 1: каждый слой сдвигает состояния токенов на один шаг затухающего потока второго порядка в обучаемом потенциале, а общее время интегрирования T = L · Δt = 8 остаётся фиксированным. Параметров — 76,7 млн.

Ключевые особенности

  • Восемь слоёв при T = 8: глубина добавляет шаги, а не ёмкость; лишние параметры — depth code, V_φ scales и register destruction gates.
  • Всего на 159 756 параметров больше SO-CODE; слои делят bank of wells, V_φ и context channels.
  • Обучение на 315 392 000 токенах; settled perplexity 5,35, лучшая — 5,26 на шаге 18 000.
  • Ниже L4-CODE (5,48) и SO-CODE (6,00), но выше matched GPT-2 (4,96); закрывает 60% разрыва с GPT-2.
  • Вторая двойка слоёв даёт около четверти эффекта первой: отдача от глубины падает.

Пример

Готового сниппета запуска в карточке нет. Основной сценарий — оценка на Python-коде: perplexity на блоке 512 по первым 2 млн токенов валидации CodeParrot. У L8-CODE settled perplexity 5,35, у L4-CODE — 5,48, у SO-CODE — 6,00.

ПодробнееБенчмарк: settled perplexity 5,35 (среднее по шагам 18 000, 18 500 и 19 000), лучшая 5,26; 76 731 865 параметров.
Обучение: 19 250 шагов за 29,5 ч в двух сессиях Colab, около 5,5 с на шаг с layer checkpointing; 315,4 млн токенов.
Ограничения: это карта, а не поток — loss минимален на обученном шаге; whole stack не проходит refinement (2,70 nats при k = 4).
С 1 400 по 3 600 шаг 34 из 45 логированных шагов имели non-finite gradient norm, обновления пропускались; причина не диагностирована. VRAM не указан.
Discuss Telegram 0 289

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # dimitarpg13/semsimula-ladder-live-code-d384-l8-none-norc-sr2
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download dimitarpg13/semsimula-ladder-live-code-d384-l8-none-norc-sr2 --local-dir ~/models/semsimula-ladder-live-code-d384-l8-none-norc-sr2
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of semsimula-ladder-live-code-d384-l8-none-norc-sr2 (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

dimitarpg13/semsimula-ladder-live-code-d384-l8-none-norc-sr2
Author
dimitarpg13
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

2890

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Language models

PM2-CODE

PM2-CODE
Language models

SR2-2B

SR2-2B