modality string | model string | device string | gpu string | embedding_dim int64 | autobatch dict | runs list |
|---|---|---|---|---|---|---|
text | sentence-transformers/all-MiniLM-L6-v2 | cuda | NVIDIA GeForce RTX 5090 | 384 | {
"min_batch": 8,
"max_batch": 512,
"start_batch": 32,
"grow_free_fraction": 0.5,
"shrink_free_fraction": 0.2,
"grow_factor": 1.5,
"shrink_factor": 0.5,
"oom_factor": 0.5
} | [
{
"num_workers": 1,
"pre_run_counts": {
"pending": 500,
"in_progress": 0,
"done": 0,
"failed": 0,
"requeued": 0
},
"total_processed": 500,
"wall_seconds": 14.76,
"throughput": 33.9,
"note": "small-scale single-worker RTX 5090 run"
}
] |
Distributed Embedding Generation Queue - Sample Embeddings
Sample text embeddings produced by a durable producer/consumer GPU queue with resume-on-crash support. Source code: github.com/narinzar/distributed-embedding-generation-queue.
Generation method
- Model:
sentence-transformers/all-MiniLM-L6-v2(384-dimensional vectors). - Pipeline: a durable SQLite task queue (WAL mode) feeds a GPU worker pool.
Items are claimed atomically, embedded in batches, written as
.npyfiles, and marked done. Orphanedin_progressitems are re-queued on restart, so a run resumes without re-embedding finished items. Batch size is tuned to GPU headroom by an autobatcher (grows with free VRAM, shrinks on a caught OOM). - Run: 500 text items embedded on an RTX 5090 at 33.9 items/s (wall 14.76s, single worker). This was a small-scale single-worker run; the architecture supports scaling the worker count, and multi-worker throughput scaling is reproducible on Linux.
Contents
sample_embeddings.npy- a 200 x 384float32array, the first 200 vectors of the 500-item run.sample_ids.txt- the item ids for those 200 vectors, one per line, aligned by row order.throughput.json- the run summary: model, device, autobatch configuration, queue transition counts, and measured throughput.
Usage
import numpy as np
from huggingface_hub import hf_hub_download
path = hf_hub_download(
repo_id="narinzar/distributed-embedding-generation-queue",
filename="sample_embeddings.npy",
repo_type="dataset",
)
vecs = np.load(path) # (200, 384) float32
License
MIT.
- Downloads last month
- 10