Allow building spin with cuda or metal support

Signed-off-by: Ryan Levick <ryan.levick@fermyon.com>

Cargo.lock

@@ -3342,12 +3342,7 @@ version = "0.2.0-dev"
 source = "git+https://github.com/rustformers/llm?rev=2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663#2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663"
 dependencies = [
  "llm-base",
- "llm-bloom",
- "llm-gpt2",
- "llm-gptj",
- "llm-gptneox",
  "llm-llama",
- "llm-mpt",
  "serde",
  "tracing",
 ]
@@ -3372,39 +3367,6 @@ dependencies = [
  "tracing",
 ]
 
-[[package]]
-name = "llm-bloom"
-version = "0.2.0-dev"
-source = "git+https://github.com/rustformers/llm?rev=2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663#2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663"
-dependencies = [
- "llm-base",
-]
-
-[[package]]
-name = "llm-gpt2"
-version = "0.2.0-dev"
-source = "git+https://github.com/rustformers/llm?rev=2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663#2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663"
-dependencies = [
- "bytemuck",
- "llm-base",
-]
-
-[[package]]
-name = "llm-gptj"
-version = "0.2.0-dev"
-source = "git+https://github.com/rustformers/llm?rev=2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663#2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663"
-dependencies = [
- "llm-base",
-]
-
-[[package]]
-name = "llm-gptneox"
-version = "0.2.0-dev"
-source = "git+https://github.com/rustformers/llm?rev=2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663#2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663"
-dependencies = [
- "llm-base",
-]
-
 [[package]]
 name = "llm-llama"
 version = "0.2.0-dev"
@@ -3414,14 +3376,6 @@ dependencies = [
  "tracing",
 ]
 
-[[package]]
-name = "llm-mpt"
-version = "0.2.0-dev"
-source = "git+https://github.com/rustformers/llm?rev=2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663#2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663"
-dependencies = [
- "llm-base",
-]
-
 [[package]]
 name = "llm-samplers"
 version = "0.0.6"

Cargo.toml

@@ -99,13 +99,11 @@ outbound-redis-tests = []
 config-provider-tests = []
 outbound-pg-tests = []
 outbound-mysql-tests = []
+llm-metal = ["spin-trigger-http/llm-metal"]
+llm-cublas = ["spin-trigger-http/llm-cublas"]
 
 [workspace]
-members = [
-  "crates/*",
-  "sdk/rust",
-  "sdk/rust/macro",
-]
+members = ["crates/*", "sdk/rust", "sdk/rust/macro"]
 
 [workspace.dependencies]
 tracing = { version = "0.1", features = ["log"] }

crates/llm/Cargo.toml

@@ -11,8 +11,9 @@ candle = { git = "https://github.com/huggingface/candle", rev = "b80348d22f8f0da
 candle-nn = { git = "https://github.com/huggingface/candle", rev = "b80348d22f8f0dadb6cc4101bde031d5de69a9a5" }
 chrono = "0.4.26"
 llm = { git = "https://github.com/rustformers/llm", rev = "2f6ffd4435799ceaa1d1bcb5a8790e5b3e0c5663", features = [
+	"tokenizers-remote",
 	"llama",
-] }
+], default-features = false }
 lru = "0.9.0"
 num_cpus = "1"
 rand = "0.8.5"
@@ -26,3 +27,8 @@ tokenizers = "0.13.4"
 tokio = { version = "1.32.0", features = ["macros", "sync"] }
 tracing = { workspace = true }
 uuid = { version = "1.4.1", features = ["v4"] }
+
+[features]
+default = []
+metal = ["llm/metal"]
+cublas = ["llm/cublas"]

crates/trigger-http/Cargo.toml

@@ -46,3 +46,8 @@ spin-testing = { path = "../testing" }
 [[bench]]
 name = "baseline"
 harness = false
+
+[features]
+default = []
+llm-metal = ["spin-trigger/llm-metal"]
+llm-cublas = ["spin-trigger/llm-cublas"]

crates/trigger/Cargo.toml

@@ -46,3 +46,8 @@ spin-componentize = { workspace = true }
 tempfile = "3.3.0"
 toml = "0.5"
 tokio = { version = "1.23", features = ["rt", "macros"] }
+
+[features]
+default = []
+llm-metal = ["spin-llm/metal"]
+llm-cublas = ["spin-llm/cublas"]