From 293d30320f5f0941ecc4fc635ee44894ed3604fc Mon Sep 17 00:00:00 2001
From: mitya <dimitry.ageev@gmail.com>
Date: Fri, 21 Jun 2024 12:29:28 +0300
Subject: [PATCH] deepseek coder v2 lite without any quantization

---
 refact_known_models/huggingface.py | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/refact_known_models/huggingface.py b/refact_known_models/huggingface.py
index e2aa6c43..cc5768fc 100644
--- a/refact_known_models/huggingface.py
+++ b/refact_known_models/huggingface.py
@@ -233,4 +233,14 @@
         "T": 8192,
         "filter_caps": ["chat"],
     },
+    "deepseek-coder-v2/16b/instruct": {
+        "backend": "transformers",
+        "model_path": "deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct",
+        "model_class_kwargs": {
+            "torch_dtype": "bf16",
+        },
+        "required_memory_mb": 80000,
+        "T": 16384,  # in fact this model can handle 128K context
+        "filter_caps": ["completion", "chat"],
+    },
 }