]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commitdiff
convert : fix Qwen3-Embedding pre-tokenizer hash (#15030)
authorDouglas Hanley <redacted>
Sat, 2 Aug 2025 10:51:02 +0000 (05:51 -0500)
committerGitHub <redacted>
Sat, 2 Aug 2025 10:51:02 +0000 (12:51 +0200)
convert_hf_to_gguf.py
convert_hf_to_gguf_update.py

index 930c1bdd025e07afff817f136ce403840b31ce90..f13f8558b2ec4a8030518ff1d66bb811a3c4293d 100755 (executable)
@@ -702,6 +702,9 @@ class TextModel(ModelBase):
         if chkhsh == "81212dc7cdb7e0c1074ca62c5aeab0d43c9f52b8a737be7b12a777c953027890":
             # ref: https://huggingface.co/moonshotai/Kimi-K2-Base
             res = "kimi-k2"
+        if chkhsh == "d4540891389ea895b53b399da6ac824becc30f2fba0e9ddbb98f92e55ca0e97c":
+            # ref: https://huggingface.co/Qwen/Qwen3-Embedding-0.6B
+            res = "qwen2"
         if chkhsh == "0ef9807a4087ebef797fc749390439009c3b9eda9ad1a097abbe738f486c01e5":
             # ref: https://huggingface.co/meta-llama/Meta-Llama-3-8B
             res = "llama-bpe"
@@ -849,9 +852,6 @@ class TextModel(ModelBase):
         if chkhsh == "2085e1638f6c377a0aa4ead21b27bb4cb941bf800df86ed391011769c1758dfb":
             # ref: https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B
             res = "exaone4"
-        if chkhsh == "d4540891389ea895b53b399da6ac824becc30f2fba0e9ddbb98f92e55ca0e97c":
-            # ref: https://huggingface.co/Qwen/Qwen3-Embedding-8B
-            res = "qwen2"
 
         if res is None:
             logger.warning("\n")
index c4904b53936f5bd198092e652901b0cfefba4be1..5e21c1f47f18c4938756957ca5e3dc9b6677f771 100755 (executable)
@@ -147,6 +147,7 @@ pre_computed_hashes = [
     {"name": "falcon-h1", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon-H1-7B-Base", "chkhsh": "3eda48b4c4dc7de733d1a8b3e3b4a85243dbbf704da2ee9d42c6beced8897896"},
     {"name": "falcon-h1", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon-H1-34B-Base", "chkhsh": "48f8e02c0359c0bbdd82f26909171fac1c18a457bb47573ed1fe3bbb2c1cfd4b"},
     {"name": "kimi-k2",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/moonshotai/Kimi-K2-Base",   "chkhsh": "81212dc7cdb7e0c1074ca62c5aeab0d43c9f52b8a737be7b12a777c953027890"},
+    {"name": "qwen2",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen3-Embedding-0.6B", "chkhsh": "d4540891389ea895b53b399da6ac824becc30f2fba0e9ddbb98f92e55ca0e97c"},
 ]