aboutsummaryrefslogtreecommitdiff
path: root/README.md
diff options
context:
space:
mode:
authorOnur Gündüz <ogunduz@gmail.com>2026-06-14 18:25:31 +0200
committerOnur Gündüz <ogunduz@gmail.com>2026-06-14 18:25:31 +0200
commit5228c61d39a7486b0a67579f4ed18222c5043da0 (patch)
tree5bf0103918114ebbacca9bf41f3bcfc9ae3301d1 /README.md
parent067ba8cc5673896f7c09866d0b9587ff7fe4b9f7 (diff)
downloadffwd-5228c61d39a7486b0a67579f4ed18222c5043da0.tar.gz
ffwd-5228c61d39a7486b0a67579f4ed18222c5043da0.zip
server: per-family embedding encoding (Qwen3 float, pplx int8)
Diffstat (limited to 'README.md')
-rw-r--r--README.md6
1 files changed, 6 insertions, 0 deletions
diff --git a/README.md b/README.md
index 8427674..715d28d 100644
--- a/README.md
+++ b/README.md
@@ -86,6 +86,12 @@ the model's output size (1024 for `Qwen3-Embedding-0.6B`, 2560 for
`Qwen3-Embedding-4B`, 4096 for `Qwen3-Embedding-8B`). Truncated Matryoshka
embeddings are re-normalized before encoding.
+`encoding_format` follows each model's API family. Qwen3-Embedding models are
+OpenAI-compatible: it defaults to `float` (the true float32 vector) and also
+accepts `base64` (base64-encoded float32). pplx-embed models default to
+`base64_int8` and also accept `base64_binary` and `float` (the decoded int8
+view). Always compare embeddings with cosine similarity.
+
Example:
```bash