Update README.md

README.md

@@ -152,17 +152,14 @@ To achieve optimal results, we recommend always including a system prompt that c
 
 ### Basic Instruct Template (V7)
 
-Without vision:
 ```
-<s>[SYSTEM_PROMPT] system prompt[/SYSTEM_PROMPT][INST] user message[/INST] assistant response</s>[INST] new user message[/INST]
+<s>[SYSTEM_PROMPT]<system prompt>[/SYSTEM_PROMPT][INST]<user message>[/INST]<assistant response></s>[INST]<user message>[/INST]
 ```
 
-With vision:
-```
-<s>[SYSTEM_PROMPT] system prompt[/SYSTEM_PROMPT][INST] user message[/INST] assistant response</s>[INST][IMG][IMG][IMG]...[IMG][IMG_BREAK][IMG]...[IMG][IMG_END] new user message[/INST]
-```
+**Be careful with subtle missing or trailing white spaces!**
+
+*Please make sure to use [mistral-common](https://github.com/mistralai/mistral-common) as the source of truth*
 
-*For more information about the tokenizer please refer to [mistral-common](https://github.com/mistralai/mistral-common)*
 
 ## Metrics
 
@@ -187,7 +184,7 @@ to implement production-ready inference pipelines with Pixtral-Large-Instruct-24
 
 **_Installation_**
 
-Make sure you install `vLLM >= v1.6.4`:
+Make sure you install `vLLM >= v0.6.4`:
 
 ```
 pip install --upgrade vllm
@@ -201,26 +198,28 @@ pip install --upgrade mistral_common
 
 You can also make use of a ready-to-go [docker image](https://github.com/vllm-project/vllm/blob/main/Dockerfile).
 
-**_Example_**
-
 ```py
 from vllm import LLM
 from vllm.sampling_params import SamplingParams
+from huggingface_hub import hf_hub_download
+from datetime import datetime, timedelta
 
-model_name = "mistralai/Pixtral-12B-2409"
-max_img_per_msg = 5
+model_name = "mistralai/Pixtral-Large-Instruct-2411"
 
-llm = LLM(model=model_name, tokenizer_mode="mistral", limit_mm_per_prompt={"image": max_img_per_msg}, max_model_len=32768)
 
 def load_system_prompt(repo_id: str, filename: str) -> str:
-    file_path = hf_hub_download(repo_id, filename)
-    with open(file_path, "r") as f:
-        prompt = f.read()
-    return prompt
+    file_path = hf_hub_download(repo_id=repo_id, filename=filename)
+    with open(file_path, 'r') as file:
+        SYSTEM_PROMPT = file.read()
+    today = datetime.today().strftime('%Y-%m-%d')
+    yesterday = (datetime.today() - timedelta(days=1)).strftime('%Y-%m-%d')
+    model_name = repo_id.split("/")[-1]
+    return SYSTEM_PROMPT.format(name=model_name, today=today, yesterday=yesterday)
 
-SYSTEM_PROMPT = load_system_prompt(model_name, "vision_system_prompt.txt")
 
-url = "https://huggingface.co/datasets/patrickvonplaten/random_img/resolve/main/yosemite.png"
+system_prompt = load_system_prompt(model_name, "SYSTEM_PROMPT.txt")
+
+user_prompt = "How many days ago was Mistral founded?"
 
 messages = [
     {
@@ -229,11 +228,18 @@ messages = [
     },
     {
         "role": "user",
-        "content": [{"type": "text", "text": "What do you see?"}, {"type": "image_url", "image_url": {"url": url}],
+        "content": prompt
     },
 ]
 
-outputs = llm.chat(messages=messages)
+sampling_params = SamplingParams(max_tokens=128_000)
+
+# note that running this model on GPU requires over 300 GB of GPU RAM
+llm = LLM(model=model_name, tokenizer_mode="mistral", tensor_parallel=8, limit_mm_per_prompt={"image": 4})
+
+outputs = llm.chat(messages, sampling_params=sampling_params)
+
+print(outputs[0].outputs[0].text)
 ```