feat: prefer disable grammar

docs/source/basic_tutorials/launcher.md

@@ -379,12 +379,12 @@ Options:
           [env: TOKENIZER_CONFIG_PATH=]
 
 ```
-## GRAMMAR_SUPPORT
+## DISABLE_GRAMMAR_SUPPORT
 ```shell
-      --grammar-support
+      --disable-grammar-support
-          Enable outlines grammar constrained generation This is a feature that allows you to generate text that follows a specific grammar
+          Disable outlines grammar constrained generation. This is a feature that allows you to generate text that follows a specific grammar
           
-          [env: GRAMMAR_SUPPORT=]
+          [env: DISABLE_GRAMMAR_SUPPORT=]
 
 ```
 ## ENV

integration-tests/conftest.py

@@ -231,7 +231,7 @@ def launcher(event_loop):
         quantize: Optional[str] = None,
         trust_remote_code: bool = False,
         use_flash_attention: bool = True,
-        grammar_support: bool = False,
+        disable_grammar_support: bool = False,
         dtype: Optional[str] = None,
     ):
         port = random.randint(8000, 10_000)
@@ -255,8 +255,8 @@ def launcher(event_loop):
 
         env = os.environ
 
-        if grammar_support:
+        if disable_grammar_support:
-            args.append("--grammar-support")
+            args.append("--disable-grammar-support")
         if num_shard is not None:
             args.extend(["--num-shard", str(num_shard)])
         if quantize is not None:
@@ -305,7 +305,7 @@ def launcher(event_loop):
         args = ["--model-id", model_id, "--env"]
 
         if grammar_support:
-            args.append("--grammar-support")
+            args.append("--disable-grammar-support")
         if num_shard is not None:
             args.extend(["--num-shard", str(num_shard)])
         if quantize is not None:

integration-tests/models/test_grammar_llama.py

@@ -7,7 +7,7 @@ from text_generation.types import GrammarType
 @pytest.fixture(scope="module")
 def flash_llama_grammar_handle(launcher):
     with launcher(
-        "TinyLlama/TinyLlama-1.1B-Chat-v1.0", num_shard=2, grammar_support=True
+        "TinyLlama/TinyLlama-1.1B-Chat-v1.0", num_shard=2, disable_grammar_support=False
     ) as handle:
         yield handle
 

launcher/src/main.rs

@@ -382,10 +382,10 @@ struct Args {
     #[clap(long, env)]
     tokenizer_config_path: Option<String>,
 
-    /// Enable outlines grammar constrained generation
+    /// Disable outlines grammar constrained generation.
     /// This is a feature that allows you to generate text that follows a specific grammar.
     #[clap(long, env)]
-    grammar_support: bool,
+    disable_grammar_support: bool,
 
     /// Display a lot of information about your runtime environment
     #[clap(long, short, action)]
@@ -1057,8 +1057,8 @@ fn spawn_webserver(
     ];
 
     // Grammar support
-    if args.grammar_support {
+    if args.disable_grammar_support {
-        router_args.push("--grammar-support".to_string());
+        router_args.push("--disable-grammar-support".to_string());
     }
 
     // Tokenizer config path

router/src/main.rs

@@ -76,7 +76,7 @@ struct Args {
     #[clap(long, env, default_value_t = false)]
     messages_api_enabled: bool,
     #[clap(long, env, default_value_t = false)]
-    grammar_support: bool,
+    disable_grammar_support: bool,
 }
 
 #[tokio::main]
@@ -110,7 +110,7 @@ async fn main() -> Result<(), RouterError> {
         ngrok_authtoken,
         ngrok_edge,
         messages_api_enabled,
-        grammar_support,
+        disable_grammar_support,
     } = args;
 
     // Launch Tokio runtime
@@ -362,7 +362,7 @@ async fn main() -> Result<(), RouterError> {
         ngrok_edge,
         tokenizer_config,
         messages_api_enabled,
-        grammar_support,
+        disable_grammar_support,
     )
     .await?;
     Ok(())

router/src/validation.rs

@@ -21,7 +21,7 @@ pub struct Validation {
     max_top_n_tokens: u32,
     max_input_length: usize,
     max_total_tokens: usize,
-    grammar_support: bool,
+    disable_grammar_support: bool,
     /// Channel to communicate with the background tokenization task
     sender: Option<mpsc::UnboundedSender<TokenizerRequest>>,
 }
@@ -35,7 +35,7 @@ impl Validation {
         max_top_n_tokens: u32,
         max_input_length: usize,
         max_total_tokens: usize,
-        grammar_support: bool,
+        disable_grammar_support: bool,
     ) -> Self {
         // If we have a fast tokenizer
         let sender = if let Some(tokenizer) = tokenizer {
@@ -70,7 +70,7 @@ impl Validation {
             max_top_n_tokens,
             max_input_length,
             max_total_tokens,
-            grammar_support,
+            disable_grammar_support,
         }
     }
 
@@ -308,7 +308,7 @@ impl Validation {
         let (grammar, grammar_type) = match grammar {
             Some(grammar) => {
                 // Ensure that grammar is not set if it's not supported
-                if !self.grammar_support {
+                if self.disable_grammar_support {
                     return Err(ValidationError::Grammar);
                 }
                 match grammar {
@@ -502,7 +502,7 @@ mod tests {
         let max_input_length = 5;
         let max_total_tokens = 6;
         let workers = 1;
-        let grammar_support = false;
+        let disable_grammar_support = true;
         let validation = Validation::new(
             workers,
             tokenizer,
@@ -511,7 +511,7 @@ mod tests {
             max_top_n_tokens,
             max_input_length,
             max_total_tokens,
-            grammar_support,
+            disable_grammar_support,
         );
 
         let max_new_tokens = 10;
@@ -532,7 +532,7 @@ mod tests {
         let max_top_n_tokens = 4;
         let max_input_length = 5;
         let max_total_tokens = 6;
-        let grammar_support = false;
+        let disable_grammar_support = true;
         let workers = 1;
         let validation = Validation::new(
             workers,
@@ -542,7 +542,7 @@ mod tests {
             max_top_n_tokens,
             max_input_length,
             max_total_tokens,
-            grammar_support,
+            disable_grammar_support,
         );
 
         let max_new_tokens = 10;
@@ -564,7 +564,7 @@ mod tests {
         let max_input_length = 5;
         let max_total_tokens = 6;
         let workers = 1;
-        let grammar_support = false;
+        let disable_grammar_support = true;
         let validation = Validation::new(
             workers,
             tokenizer,
@@ -573,7 +573,7 @@ mod tests {
             max_top_n_tokens,
             max_input_length,
             max_total_tokens,
-            grammar_support,
+            disable_grammar_support,
         );
         match validation
             .validate(GenerateRequest {
@@ -600,7 +600,7 @@ mod tests {
         let max_input_length = 5;
         let max_total_tokens = 106;
         let workers = 1;
-        let grammar_support = false;
+        let disable_grammar_support = true;
         let validation = Validation::new(
             workers,
             tokenizer,
@@ -609,7 +609,7 @@ mod tests {
             max_top_n_tokens,
             max_input_length,
             max_total_tokens,
-            grammar_support,
+            disable_grammar_support,
         );
         match validation
             .validate(GenerateRequest {
@@ -665,7 +665,7 @@ mod tests {
         let max_input_length = 5;
         let max_total_tokens = 106;
         let workers = 1;
-        let grammar_support = false;
+        let disable_grammar_support = true;
         let validation = Validation::new(
             workers,
             tokenizer,
@@ -674,7 +674,7 @@ mod tests {
             max_top_n_tokens,
             max_input_length,
             max_total_tokens,
-            grammar_support,
+            disable_grammar_support,
         );
         match validation
             .validate(GenerateRequest {