Idefics2: sync added image tokens with transformers

Before this change, the number of reserved image tokens was not the
same as the number of images. Fixes #2029.

While at it, also remove all the image token handling duplication
in `prepare_input`.

README.md

@@ -153,7 +153,7 @@ this will impact performance.
 ### Distributed Tracing
 
 `text-generation-inference` is instrumented with distributed tracing using OpenTelemetry. You can use this feature
-by setting the address to an OTLP collector with the `--otlp-endpoint` argument. The default service name can be 
+by setting the address to an OTLP collector with the `--otlp-endpoint` argument. The default service name can be
 overridden with the `--otlp-service-name` argument
 
 ### Architecture

benchmark/src/main.rs

@@ -147,7 +147,9 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
             tracing::info!("Downloading tokenizer");
 
             // Parse Huggingface hub token
-            let auth_token = std::env::var("HF_TOKEN").or_else(|_| std::env::var("HUGGING_FACE_HUB_TOKEN")).ok();
+            let auth_token = std::env::var("HF_TOKEN")
+                .or_else(|_| std::env::var("HUGGING_FACE_HUB_TOKEN"))
+                .ok();
 
             // Download and instantiate tokenizer
             // We need to download it outside of the Tokio runtime

clients/python/text_generation/types.py

@@ -455,6 +455,6 @@ class DeployedModel(BaseModel):
     # Disable warning for use of `model_` prefix in `model_id`. Be mindful about adding members
     # with model_ prefixes, since this disables guardrails for colliding fields:
     # https://github.com/pydantic/pydantic/issues/9177
-    model_config  = ConfigDict(protected_namespaces=())
+    model_config = ConfigDict(protected_namespaces=())
     model_id: str
     sha: str