Update README.md

README.md

@@ -4,8 +4,80 @@ language:
 pipeline_tag: text-generation
 tags:
 - llama-2
+- function calling
 license: llama2
 ---
+# **XAgent Llama**
+
+XAgentLlaMa is a collection of fine-tuned generative text models ranging in scale from 7 billion to 34 billion based on Llama 2 and Code Llama.
+This is the repository for the 34B fine-tuned model, optimized for XAgent with strong function call ability.
+
+## Warning: This is a preview version of the model, does not stand for final quality.
+
+We collect around 300K pieces of data and fine-tune Code-Llama 34B with 48 A100 GPUs. More details will be released later.
+
+This model is trained with a special function call format, and should be used with [XAgentGen](https://github.com/OpenBMB/XAgent/tree/dev/XAgentGen) to get best performance.
+
+### XAgentGen input format:
+```json
+"messages":[
+    {
+    "role":"system",
+    "content":"...."
+    },
+    {...}
+],
+"global_arguments":{ // Optional
+    "type": "object",
+    "properties":{
+        "k1":{
+            "type":"integer",
+            "description":"..."
+        },
+        "k2":{
+            "type":"string",
+            "description":"..."
+        },
+        ...
+    },
+    "required":["k1","k2"]
+},
+"functions":[// Optional
+    {
+        "name":"func1",
+        "description":"...",
+        "parameters":{
+            "type":"object",
+            "properties":{...},
+            "required":[...]
+        }
+    },
+    ....
+],
+"function_call": {// Optional
+    "name":"func1"
+}
+```
+
+### XAgentGen call output format:
+```json
+{
+    "global_arguments": {
+        "k1":"v1",
+        "k2":"v2",
+        "k3":"v3",
+        ...
+    },
+    "function_call":{
+        "name":"func1",
+        "arguments":{...}
+    }
+}
+```
+If the json format of `global_arguments` is provided, the output will contains the `global_arguments` at any time.
+
+
+
 # **Code Llama**
 Code Llama is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 34 billion parameters. This is the repository for the 7B instruct-tuned version in the Hugging Face Transformers format. This model is designed for general code synthesis and understanding. Links to other models can be found in the index at the bottom.