feat: Qwen2 (#1608)

See #1584 --------- Co-authored-by: Cheng Kuan Yong Jason <[email protected]>
huggingface · Feb 28, 2024 · 26cdea5 · 26cdea5
1 parent b40e833
commit 26cdea5
Show file tree

Hide file tree

Showing 17 changed files with 1,159 additions and 24 deletions.
diff --git a/integration-tests/models/__snapshots__/test_flash_qwen2/test_flash_qwen2.json b/integration-tests/models/__snapshots__/test_flash_qwen2/test_flash_qwen2.json
@@ -0,0 +1,84 @@
+{
+  "details": {
+    "best_of_sequences": null,
+    "finish_reason": "length",
+    "generated_tokens": 10,
+    "prefill": [
+      {
+        "id": 2271,
+        "logprob": null,
+        "text": "Test"
+      },
+      {
+        "id": 1681,
+        "logprob": -8.8515625,
+        "text": " request"
+      }
+    ],
+    "seed": null,
+    "tokens": [
+      {
+        "id": 198,
+        "logprob": -2.9023438,
+        "special": false,
+        "text": "\n"
+      },
+      {
+        "id": 2,
+        "logprob": -2.9160156,
+        "special": false,
+        "text": "#"
+      },
+      {
+        "id": 4230,
+        "logprob": -3.1035156,
+        "special": false,
+        "text": " Create"
+      },
+      {
+        "id": 264,
+        "logprob": -1.1025391,
+        "special": false,
+        "text": " a"
+      },
+      {
+        "id": 1681,
+        "logprob": -1.6914062,
+        "special": false,
+        "text": " request"
+      },
+      {
+        "id": 198,
+        "logprob": -1.1953125,
+        "special": false,
+        "text": "\n"
+      },
+      {
+        "id": 2035,
+        "logprob": -1.3203125,
+        "special": false,
+        "text": "request"
+      },
+      {
+        "id": 284,
+        "logprob": -0.13537598,
+        "special": false,
+        "text": " ="
+      },
+      {
+        "id": 7388,
+        "logprob": -1.2402344,
+        "special": false,
+        "text": " requests"
+      },
+      {
+        "id": 670,
+        "logprob": -0.2775879,
+        "special": false,
+        "text": ".get"
+      }
+    ],
+    "top_tokens": null
+  },
+  "generated_text": "\n# Create a request\nrequest = requests.get"
+}
diff --git a/integration-tests/models/__snapshots__/test_flash_qwen2/test_flash_qwen2_all_params.json b/integration-tests/models/__snapshots__/test_flash_qwen2/test_flash_qwen2_all_params.json
@@ -0,0 +1,84 @@
+{
+  "details": {
+    "best_of_sequences": null,
+    "finish_reason": "length",
+    "generated_tokens": 10,
+    "prefill": [
+      {
+        "id": 2271,
+        "logprob": null,
+        "text": "Test"
+      },
+      {
+        "id": 1681,
+        "logprob": -8.8515625,
+        "text": " request"
+      }
+    ],
+    "seed": 0,
+    "tokens": [
+      {
+        "id": 311,
+        "logprob": -1.4277344,
+        "special": false,
+        "text": " to"
+      },
+      {
+        "id": 279,
+        "logprob": -0.65478516,
+        "special": false,
+        "text": " the"
+      },
+      {
+        "id": 2473,
+        "logprob": -1.8300781,
+        "special": false,
+        "text": " service"
+      },
+      {
+        "id": 382,
+        "logprob": -0.75,
+        "special": false,
+        "text": ".\n\n"
+      },
+      {
+        "id": 286,
+        "logprob": -0.11621094,
+        "special": false,
+        "text": "       "
+      },
+      {
+        "id": 549,
+        "logprob": 0.0,
+        "special": false,
+        "text": " :"
+      },
+      {
+        "id": 689,
+        "logprob": -0.48608398,
+        "special": false,
+        "text": "return"
+      },
+      {
+        "id": 25,
+        "logprob": 0.0,
+        "special": false,
+        "text": ":"
+      },
+      {
+        "id": 5949,
+        "logprob": -0.5756836,
+        "special": false,
+        "text": " Response"
+      },
+      {
+        "id": 504,
+        "logprob": -0.24499512,
+        "special": false,
+        "text": " from"
+      }
+    ],
+    "top_tokens": null
+  },
+  "generated_text": "Test request to the service.\n\n        :return: Response from"
+}