A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 00:45 UTC
[Doc] Replace ibm-fms with ibm-ai-platform (#12709)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
Thomas Parnell committed
Feb 4, 2025 at 02:05 UTC
bb392af434a49b3f8655f0e78737ced6524056b7
5 files changed
+10
-10
docs/source/features/spec_decode.md
+6
-6
@@ -131,7 +131,7 @@ sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
131
llm = LLM(
132
model="meta-llama/Meta-Llama-3.1-70B-Instruct",
133
tensor_parallel_size=4,
134
- speculative_model="ibm-fms/llama3-70b-accelerator",
134
+ speculative_model="ibm-ai-platform/llama3-70b-accelerator",
135
speculative_draft_tensor_parallel_size=1,
136
)
137
outputs = llm.generate(prompts, sampling_params)
@@ -149,11 +149,11 @@ limitation will be fixed in a future release.
149
150
A variety of speculative models of this type are available on HF hub:
151
152
-- [llama-13b-accelerator](https://huggingface.co/ibm-fms/llama-13b-accelerator)
153
-- [llama3-8b-accelerator](https://huggingface.co/ibm-fms/llama3-8b-accelerator)
154
-- [codellama-34b-accelerator](https://huggingface.co/ibm-fms/codellama-34b-accelerator)
155
-- [llama2-70b-accelerator](https://huggingface.co/ibm-fms/llama2-70b-accelerator)
156
-- [llama3-70b-accelerator](https://huggingface.co/ibm-fms/llama3-70b-accelerator)
152
+- [llama-13b-accelerator](https://huggingface.co/ibm-ai-platform/llama-13b-accelerator)
153
+- [llama3-8b-accelerator](https://huggingface.co/ibm-ai-platform/llama3-8b-accelerator)
154
+- [codellama-34b-accelerator](https://huggingface.co/ibm-ai-platform/codellama-34b-accelerator)
155
+- [llama2-70b-accelerator](https://huggingface.co/ibm-ai-platform/llama2-70b-accelerator)
156
+- [llama3-70b-accelerator](https://huggingface.co/ibm-ai-platform/llama3-70b-accelerator)
157
- [granite-3b-code-instruct-accelerator](https://huggingface.co/ibm-granite/granite-3b-code-instruct-accelerator)
158
- [granite-8b-code-instruct-accelerator](https://huggingface.co/ibm-granite/granite-8b-code-instruct-accelerator)
159
- [granite-7b-instruct-accelerator](https://huggingface.co/ibm-granite/granite-7b-instruct-accelerator)
examples/offline_inference/mlpspeculator.py
+1
-1
@@ -51,7 +51,7 @@ if __name__ == "__main__":
51
# Create an LLM with spec decoding
52
llm = LLM(
53
model="meta-llama/Llama-2-13b-chat-hf",
54
- speculative_model="ibm-fms/llama-13b-accelerator",
54
+ speculative_model="ibm-ai-platform/llama-13b-accelerator",
55
)
56
57
print("With speculation")
tests/models/registry.py
+1
-1
@@ -278,7 +278,7 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
278
"MedusaModel": _HfExamplesInfo("JackFram/llama-68m",
279
speculative_model="abhigoyal/vllm-medusa-llama-68m-random"), # noqa: E501
280
"MLPSpeculatorPreTrainedModel": _HfExamplesInfo("JackFram/llama-160m",
281
- speculative_model="ibm-fms/llama-160m-accelerator"), # noqa: E501
281
+ speculative_model="ibm-ai-platform/llama-160m-accelerator"), # noqa: E501
282
}
283
284
_FALLBACK_MODEL = {
tests/spec_decode/e2e/test_mlp_correctness.py
+1
-1
@@ -33,7 +33,7 @@ from .conftest import run_equality_correctness_test
33
MAIN_MODEL = "JackFram/llama-160m"
34
35
# speculative model
36
-SPEC_MODEL = "ibm-fms/llama-160m-accelerator"
36
+SPEC_MODEL = "ibm-ai-platform/llama-160m-accelerator"
37
38
# max. number of speculative tokens: this corresponds to
39
# n_predict in the config.json of the speculator model.
vllm/model_executor/models/mlp_speculator.py
+1
-1
@@ -64,7 +64,7 @@ class MLPSpeculator(nn.Module):
64
https://arxiv.org/pdf/2404.19124
65
66
Trained speculators of this type are available on HF hub at:
67
- https://huggingface.co/ibm-fms and https://huggingface.co/ibm-granite
67
+ https://huggingface.co/ibm-ai-platform and https://huggingface.co/ibm-granite
68
"""
69
70
def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None: