Primary use stated or inferred cautiously from official documentation.
B
BYTEDANCE SEED / ByteDance-Seed/Cross-Layer-TopK-CLT
Cross-Layer TopK CLT
ByteDance Seed sparse transcoder for studying Qwen3-8B-Base. It reads the residual stream before the MLPs and approximates their outputs using features across layers. The verified revision contains all 36 encoders and 36 decoders; uploading finished on November 5.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | ByteDance-Seed/Cross-Layer-TopK-CLT |
|---|---|
| Model type | Interpretability |
| Access model | Open weights |
| License | Apache 2.0 |
| Deployment | Self-hosted |
| Release | 05 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Internal activations |
| Exit | Latent features · Reconstructed activations |
| Context window | 36 layers and 32,768 features per layer |
| maximum output | 360 active features per token across layers |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · PyTorch |
Analyze activations and reconstruct layer contributions for a fixed base model. This is a research component: it does not accept text as a chat API, does not generate replies, and its internal features are not search embeddings.
The full set occupies about 377 GB in FP32 and contains 94.2 billion parameters. Configuration and scale factors must match the revision. The encoding-only example omits decoders and does not show that the complete model fits on one GPU.
03 / CAPABILITIES
What it can do
Orientation
Analyze activations and reconstruct layer contributions for a fixed base model. This is a research component: it does not accept text as a chat API, does not generate replies, and its internal features are not search embeddings.
Context
36 layers and 32,768 features per layer · 360 active features per token across layers
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Input modalities
Internal activations
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
No hosted commercial API has been verified for this checkpoint. Downloadable weights do not cover storage, accelerators, memory or operating costs.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
The card documents encoding and reconstruction without a verified numeric score for this checkpoint. A separate evaluation should pin the base model and measure reconstruction error and ablation sensitivity. File size and sparsity are not quality metrics.
i No verified numeric score for the exact version.
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Cross-Layer TopK CLT
Version added to Inferama's verified catalog.
07 / SOURCES