Upload folder using huggingface_hub

README.md

@@ -0,0 +1,36 @@
+---
+language:
+- en
+- km
+license: mit
+library_name: netra-nmt
+pipeline_tag: translation
+tags:
+- translation
+- khmer
+- english
+- diffusion
+- non-autoregressive
+---
+
+# netra-nmt (diffusion, experimental)
+
+A non-autoregressive **discrete-diffusion** decoder variant of netra-nmt, same
+encoder-decoder size (~90M params) as the autoregressive model, trading some
+translation quality for fixed-cost, parallel CPU decoding (see the main repo
+README for the quality/latency comparison).
+
+## Usage
+
+```bash
+pip install netra-nmt
+```
+
+```python
+from netra_nmt import DiffusionTranslator
+t = DiffusionTranslator(repo_id="Darayut/netra-nmt-small-diffusion", steps=10)
+print(t.translate("Hello, how are you?", direction="en2km"))
+print(t.translate("សួស្តី", direction="km2en"))
+```
+
+Files: `model.safetensors` (fp16 weights), `config.json`, `spm_32k.model` (SentencePiece tokenizer).

config.json

@@ -0,0 +1,20 @@
+{
+  "d_model": 512,
+  "enc_layers": 6,
+  "dec_layers": 6,
+  "n_heads": 8,
+  "ffn_dim": 2048,
+  "max_len": 256,
+  "pad_id": 0,
+  "bos_id": 2,
+  "eos_id": 3,
+  "vocab_size": 32000,
+  "mask_id": 32000,
+  "decode_steps": 10,
+  "lang_markers": {
+    "en": "<2en>",
+    "km": "<2km>"
+  },
+  "dtype": "float16",
+  "model_type": "diffusion"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa5fec147fa7acb6fd92708734959292b30022b3f2a90b4fb9b3698ae6d1d0ac
+size 179726720

spm_32k.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8daf6a441688a671e97cd579de636e86870d3bf4c3796c6f3a2e39dd8363051
+size 1052926