Update the candle and cudarc versions.

kyutai-labs · Oct 2, 2024 · 967b7f2 · 967b7f2
1 parent 1700456
commit 967b7f2
Show file tree

Hide file tree

Showing 4 changed files with 7 additions and 8 deletions.
diff --git a/rust/Cargo.toml b/rust/Cargo.toml
@@ -18,10 +18,10 @@ categories = ["science"]
 
 
 [workspace.dependencies]
-candle = { version = "0.7.1", package = "candle-core" }
-candle-nn = "0.7.1"
-candle-transformers = "0.7.1"
-candle-flash-attn = "0.7.1"
+candle = { version = "0.7.2", package = "candle-core" }
+candle-nn = "0.7.2"
+candle-transformers = "0.7.2"
+candle-flash-attn = "0.7.2"
 
 [profile.release]
 debug = true

diff --git a/rust/moshi-backend/config-q8.json b/rust/moshi-backend/config-q8.json
@@ -1,7 +1,7 @@
 {
   "instance_name": "foo",
   "hf_repo": "kyutai/moshiko-candle-q8",
-  "lm_model_file": "$HOME/tmp/model.q8.gguf",
+  "lm_model_file": "$HOME/tmp/moshiko_rs_301e30bf@120.q8.gguf",
   "text_tokenizer_file": "$HOME/tmp/tokenizer_spm_32k_3.model",
   "log_dir": "$HOME/tmp/moshi-logs",
   "encodec_model_file": "$HOME/tmp/tokenizer-e351c8d8-checkpoint125.safetensors",

diff --git a/rust/moshi-backend/config.json b/rust/moshi-backend/config.json
@@ -1,7 +1,7 @@
 {
   "instance_name": "foo",
   "hf_repo": "kyutai/moshiko-candle-bf16",
-  "lm_model_file": "$HOME/tmp/model.safetensors",
+  "lm_model_file": "$HOME/tmp/moshiko_rs_301e30bf@120.safetensors",
   "text_tokenizer_file": "$HOME/tmp/tokenizer_spm_32k_3.model",
   "log_dir": "$HOME/tmp/moshi-logs",
   "encodec_model_file": "$HOME/tmp/tokenizer-e351c8d8-checkpoint125.safetensors",

diff --git a/rust/moshi-core/Cargo.toml b/rust/moshi-core/Cargo.toml
@@ -14,14 +14,13 @@ candle = { workspace = true }
 candle-nn = { workspace = true }
 candle-transformers = { workspace = true }
 candle-flash-attn = { workspace = true, optional = true }
-cudarc = { version = "=0.11.6", features = ["std", "cublas", "cublaslt", "curand", "driver", "nvrtc", "f16", "cuda-version-from-build-system", "dynamic-linking"], default-features=false, optional = true }
 
 rayon = "1.8.1"
 serde = { version = "1.0", features = ["derive"] }
 tracing = "0.1.40"
 
 [features]
 default = []
-cuda = ["candle/cuda", "candle-nn/cuda", "cudarc"]
+cuda = ["candle/cuda", "candle-nn/cuda"]
 metal = ["candle/metal", "candle-nn/metal"]
 flash-attn = ["cuda", "dep:candle-flash-attn"]