feat: Use the new v2 memory crate

Signed-off-by: Alexandre Milesi <[email protected]>

Author: milesial

Cargo.lock

@@ -47,6 +47,7 @@ keywords = ["llm", "genai", "inference", "nvidia", "distributed"]
 dynamo-runtime = { path = "lib/runtime", version = "0.6.1" }
 dynamo-llm = { path = "lib/llm", version = "0.6.1" }
 dynamo-config = { path = "lib/config", version = "0.6.1" }
+dynamo-memory = { path = "lib/memory", version = "0.6.1" }
 dynamo-tokens = { path = "lib/tokens", version = "0.6.1" }
 dynamo-async-openai = { path = "lib/async-openai", version = "0.6.1", features = [
     "byot",

Cargo.toml

@@ -13,7 +13,7 @@ readme.workspace = true
 description = "Dynamo LLM Library"
 
 [features]
-default = ["block-manager"]
+default = []
 # todo(ops): get this working in CI as a default.
 # default = ["block-manager", "testing-full"]
 
@@ -33,6 +33,7 @@ harness = false
 name = "transfer_context_v2"
 harness = false
 required-features = ["block-manager", "testing-cuda"]
+
 [dependencies]
 # repo
 dynamo-runtime = { workspace = true }
@@ -41,6 +42,7 @@ dynamo-runtime = { workspace = true }
 aho-corasick = "1.1"
 anyhow = { workspace = true }
 dynamo-async-openai = { workspace = true }
+dynamo-memory = { workspace = true }
 dynamo-parsers = { workspace = true }
 async-stream = { workspace = true }
 async-trait = { workspace = true }

lib/bindings/python/Cargo.lock

@@ -11,7 +11,7 @@ use dynamo_async_openai::types::ChatCompletionRequestUserMessageContentPart;
 use super::common::EncodedMediaData;
 use super::decoders::{Decoder, MediaDecoder};
 use super::rdma::{RdmaMediaDataDescriptor, get_nixl_agent};
-use nixl_sys::Agent as NixlAgent;
+use dynamo_memory::nixl::NixlAgent;
 
 const DEFAULT_HTTP_USER_AGENT: &str = "dynamo-ai/dynamo";
 const DEFAULT_HTTP_TIMEOUT: Duration = Duration::from_secs(30);