feat: update docs refactor and avoid cuda graphs when unavailable

clients/python/text_generation/client.py

@@ -118,6 +118,9 @@ class Client:
                 Return the decoder input token logprobs and ids
             top_n_tokens (`int`):
                 Return the `n` most likely tokens at each step
+            grammar (`Grammar`):
+                Whether to use a grammar for the generation and the grammar to use. Grammars will constrain the generation
+                of the text to match a regular expression or JSON schema.
 
         Returns:
             Response: generated response
@@ -209,6 +212,9 @@ class Client:
                 Watermarking with [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226)
             top_n_tokens (`int`):
                 Return the `n` most likely tokens at each step
+            grammar (`Grammar`):
+                Whether to use a grammar for the generation and the grammar to use. Grammars will constrain the generation
+                of the text to match a regular expression or JSON schema.
 
         Returns:
             Iterator[StreamResponse]: stream of generated tokens
@@ -372,6 +378,9 @@ class AsyncClient:
                 Return the decoder input token logprobs and ids
             top_n_tokens (`int`):
                 Return the `n` most likely tokens at each step
+            grammar (`Grammar`):
+                Whether to use a grammar for the generation and the grammar to use. Grammars will constrain the generation
+                of the text to match a regular expression or JSON schema.
 
         Returns:
             Response: generated response
@@ -462,6 +471,9 @@ class AsyncClient:
                 Watermarking with [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226)
             top_n_tokens (`int`):
                 Return the `n` most likely tokens at each step
+            grammar (`Grammar`):
+                Whether to use a grammar for the generation and the grammar to use. Grammars will constrain the generation
+                of the text to match a regular expression or JSON schema.
 
         Returns:
             AsyncIterator[StreamResponse]: stream of generated tokens

router/src/infer.rs

@@ -198,13 +198,12 @@ impl Infer {
     pub(crate) fn apply_completion_template(
         &self,
         prompt: String,
-        prefix: Option<String>,
         suffix: Option<String>,
     ) -> Result<String, InferError> {
         self.completion_template
             .as_ref()
             .ok_or_else(|| InferError::TemplateError(ErrorKind::TemplateNotFound.into()))?
-            .apply(prompt, prefix, suffix)
+            .apply(prompt, suffix)
             .map_err(|e| {
                 metrics::increment_counter!("tgi_request_failure", "err" => "template");
                 tracing::error!("{e}");
@@ -386,15 +385,9 @@ impl CompletionTemplate {
         Self { template }
     }
 
-    fn apply(
-        &self,
-        prompt: String,
-        prefix: Option<String>,
-        suffix: Option<String>,
-    ) -> Result<String, InferError> {
+    fn apply(&self, prompt: String, suffix: Option<String>) -> Result<String, InferError> {
         self.template
             .render(CompletionTemplateInputs {
-                prefix: prefix.as_deref(),
                 prompt: prompt.as_str(),
                 suffix: suffix.as_deref(),
             })

router/src/lib.rs

@@ -595,7 +595,6 @@ pub(crate) struct ChatTemplateInputs<'a> {
 
 #[derive(Clone, Serialize, Deserialize)]
 pub(crate) struct CompletionTemplateInputs<'a> {
-    prefix: Option<&'a str>,
     prompt: &'a str,
     suffix: Option<&'a str>,
 }

router/src/server.rs

@@ -578,7 +578,7 @@ async fn completions(
     let stream = req.stream.unwrap_or_default();
     let seed = req.seed;
 
-    let inputs = match infer.apply_completion_template(req.prompt, None, req.suffix) {
+    let inputs = match infer.apply_completion_template(req.prompt, req.suffix) {
         Ok(inputs) => inputs,
         Err(err) => {
             metrics::increment_counter!("tgi_request_failure", "err" => "validation");
@@ -1263,8 +1263,8 @@ pub async fn run(
         .route("/generate", post(generate))
         .route("/generate_stream", post(generate_stream))
         .route("/v1/chat/completions", post(chat_completions))
-        .route("/vertex", post(vertex_compatibility))
         .route("/v1/completions", post(completions))
+        .route("/vertex", post(vertex_compatibility))
         .route("/tokenize", post(tokenize))
         .route("/health", get(health))
         .route("/ping", get(health))

server/text_generation_server/models/__init__.py

@@ -8,7 +8,6 @@ from typing import Optional
 from text_generation_server.utils.speculate import get_speculate, set_speculate
 from text_generation_server.models.model import Model
 from text_generation_server.models.causal_lm import CausalLM
-from text_generation_server.models.flash_causal_lm import FlashCausalLM
 from text_generation_server.models.bloom import BLOOMSharded
 from text_generation_server.models.mpt import MPTSharded
 from text_generation_server.models.seq2seq_lm import Seq2SeqLM
@@ -34,7 +33,6 @@ __all__ = [
     "Model",
     "BLOOMSharded",
     "CausalLM",
-    "FlashCausalLM",
     "GalacticaSharded",
     "Seq2SeqLM",
     "SantaCoder",
@@ -45,7 +43,20 @@ __all__ = [
 
 FLASH_ATT_ERROR_MESSAGE = "{} requires Flash Attention enabled models."
 
-FLASH_ATTENTION = True
+# FlashCausalLM reqiures CUDA Graphs to be enabled on the system. This will throw a RuntimeError
+# if CUDA Graphs are not available when calling `torch.cuda.graph_pool_handle()` in the FlashCausalLM
+HAS_CUDA_GRAPH = False
+try:
+    from text_generation_server.models.flash_causal_lm import FlashCausalLM
+
+    HAS_CUDA_GRAPH = True
+except RuntimeError as e:
+    logger.warning(f"Could not import FlashCausalLM: {e}")
+
+if HAS_CUDA_GRAPH:
+    __all__.append(FlashCausalLM)
+
+
 try:
     from text_generation_server.models.flash_rw import FlashRWSharded
     from text_generation_server.models.flash_neox import FlashNeoXSharded