tb-labs /
RuVision-7B-Instruct
Vision-language model for Russian-language image QA, OCR-aware captioning and document understanding.
v2.4 · resumable chunked uploads up to 5 TB
Private storage and distribution of heavy model weights and datasets for data-science teams. Version every checkpoint, ship .safetensors at line speed, and keep everything inside your security perimeter.
$ pip install tensorbase-cli
# authenticate with a workspace token
$ tb login --token $TB_TOKEN
# push a checkpoint (chunked, resumable)
$ tb push ./ckpt acme/RuVision-7B-Instruct
model-00001-of-00003.safetensors 4.98 GB ████████████ 100%
model-00002-of-00003.safetensors 4.97 GB ████████████ 100%
model-00003-of-00003.safetensors 4.25 GB ███████░░░░░ 61%
# pull anywhere, verified by sha256
$ tb pull acme/RuVision-7B-Instruct@a91f3c0
Most pulled public checkpoints this week.
tb-labs /
Vision-language model for Russian-language image QA, OCR-aware captioning and document understanding.
volga-ai /
Speech recognition fine-tuned on 12k hours of Russian call-center and broadcast audio. WER 6.8% on internal test set.
finlab /
Instruction-tuned LLM for financial reporting, IFRS / RAS terminology and regulatory Q&A. 8k context.
01 / storage
Files are split into 64 MiB chunks keyed by SHA-256. Unchanged layers are never uploaded twice.
02 / transfer
HTTP Range requests and parallel streams saturate 10 Gbit/s links. Interrupted transfers continue where they stopped.
03 / access
SAML 2.0 / OIDC, SCIM provisioning, scoped read-only tokens for CI runners and inference clusters.
04 / compliance
Storage in Moscow and Saint Petersburg data centers, full audit log, on-premise edition for closed networks.
Free tier for researchers. Team and Business plans from 2 490 ₽ per seat.