Update README.md

README.md

@@ -13,6 +13,19 @@ tags:
 ---
 # Model Card — logx v0.1.0
 
+
+Run directly with 🤗 Transformers:
+
+```python
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+tok = AutoTokenizer.from_pretrained("path/to/unzipped/model")
+model = AutoModelForSeq2SeqLM.from_pretrained("path/to/unzipped/model")
+ids = tok("parse: how many 502s in the last hour", return_tensors="pt").input_ids
+print(tok.decode(model.generate(ids, max_length=128)[0], skip_special_tokens=True))
+# {"action":"count","source":"nginx_access","filters":[{"field":"status","op":"eq","value":"502"}],"time":{"last":"1h"}}
+```
+
+
 NL → nginx-log-query translation model. Fine-tune of
 [`google/t5-efficient-tiny`](https://huggingface.co/google/t5-efficient-tiny)
 (15.57M params, Apache-2.0) that maps English questions about nginx logs to a
@@ -38,17 +51,6 @@ git clone https://github.com/devang007/logX && cd logX
 logx -q "top 5 ips" -src /var/log/nginx/access.log
 ```
 
-Or directly with 🤗 Transformers:
-
-```python
-from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
-tok = AutoTokenizer.from_pretrained("path/to/unzipped/model")
-model = AutoModelForSeq2SeqLM.from_pretrained("path/to/unzipped/model")
-ids = tok("parse: how many 502s in the last hour", return_tensors="pt").input_ids
-print(tok.decode(model.generate(ids, max_length=128)[0], skip_special_tokens=True))
-# {"action":"count","source":"nginx_access","filters":[{"field":"status","op":"eq","value":"502"}],"time":{"last":"1h"}}
-```
-
 The tokenizer is **not** a stock T5 tokenizer: 6 tokens (`< \ ^ { } ~`) were
 added so JSON braces survive encoding, with a patched Metaspace pre-tokenizer.
 Always load the tokenizer shipped in the zip, never the base model's.