feat(sfttrainer): support granite-3.3-8b (#276)

Signed-off-by: Vassilis Vassiliadis <[email protected]>

Author: VassilisVassiliadis

plugins/actuators/sfttrainer/ado_actuators/sfttrainer/README.md

@@ -233,7 +233,7 @@ Required:
 
 - model_name: Supported models:
   <!-- markdownlint-disable-next-line line-length -->
-  `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]`
+  `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny", "granite-3.3-8b"]`
 - model_max_length: Maximum sequence length. Sequences will be right padded (and
   possibly truncated)
 - number_gpus: The effective number of GPUs (to be evenly distributed to
@@ -610,7 +610,7 @@ Versioning:
 
 - model_name: Supported models:
   <!-- markdownlint-disable-next-line line-length -->
-  `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]`
+  `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny", "granite-3.3-8b"]`
 - dataset_id: One of
   - `news-chars-512-entries-4096`: 4096 entries with samples of 512 + 127
     (prompt) + 512 characters
@@ -862,7 +862,7 @@ Required:
 
 - model_name: Supported models:
   <!-- markdownlint-disable-next-line line-length -->
-  `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]`
+  `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny", "granite-3.3-8b"]`
 - model_max_length: Maximum sequence length. Sequences will be right padded (and
   possibly truncated)
 - number_gpus: The effective number of GPUs (to be evenly distributed to
@@ -1028,6 +1028,7 @@ Sets the `--target_modules` layer names based on the `model_name`:
 - `mistral-123b-v2`: `["q_proj", "v_proj"]`
 - `mixtral-8x7b-instruct-v0.1`: `["q_proj", "v_proj"]`
 - `granite-3-8b`: `["q_proj", "v_proj"]`
+- `granite-3.3-8b`: `["q_proj", "v_proj"]`
 - `granite-3.1-2b`: `["q_proj", "v_proj"]`
 - `granite-3.1-8b-instruct`: `["q_proj", "v_proj"]`
 - `llava-v1.6-mistral-7b`: `["q_proj", "v_proj"]`
@@ -1725,7 +1726,7 @@ Required:
 
 - model_name: Supported models:
   <!-- markdownlint-disable-next-line line-length -->
-  `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]`
+  `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny", "granite-3.3-8b"]`
 - model_max_length: Maximum sequence length. Sequences will be right padded (and
   possibly truncated)
 - number_gpus: The effective number of GPUs (to be evenly distributed to

plugins/actuators/sfttrainer/ado_actuators/sfttrainer/config/models.yaml

@@ -10,6 +10,8 @@ granite-20b-v2:
   Vanilla: /hf-models-pvc/granite-20b-code-base-v2/step_280000_ckpt/
 granite-3-8b:
   Vanilla: ibm-granite/granite-3.0-8b-base
+granite-3.3-8b:
+  Vanilla: ibm-granite/granite-3.3-8b-base
 granite-3.1-2b:
   Vanilla: ibm-granite/granite-3.1-2b-base
 granite-3.1-3b-a800m-instruct:

plugins/actuators/sfttrainer/ado_actuators/sfttrainer/experiments/lora.py

@@ -49,6 +49,7 @@
     "granite-3.1-3b-a800m-instruct": ["q_proj", "v_proj"],
     "granite-vision-3.2-2b": ["q_proj", "v_proj"],
     "llava-v1.6-mistral-7b": ["q_proj", "v_proj"],
+    "granite-3.3-8b": ["q_proj", "v_proj"],
 }
 
 

website/docs/actuators/sft-trainer.md

@@ -207,7 +207,7 @@ models.
 
     - model_name: Supported models:
       <!-- markdownlint-disable-next-line line-length -->
-      `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]`
+      `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny", "granite-3.3-8b"]`
     - model_max_length: Maximum sequence length. Sequences will be right padded
       (and possibly truncated)
     - number_gpus: The effective number of GPUs (to be evenly distributed to
@@ -469,7 +469,7 @@ configurations.
 
     - model_name: Supported models:
       <!-- markdownlint-disable-next-line line-length -->
-      `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]`
+      `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny", "granite-3.3-8b"]`
     - model_max_length: Maximum sequence length. Sequences will be right padded (and
       possibly truncated)
     - number_gpus: The effective number of GPUs (to be evenly distributed to
@@ -754,7 +754,7 @@ adaptation.
 
     - model_name: Supported models:
       <!-- markdownlint-disable-next-line line-length -->
-      `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]`
+      `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny", "granite-3.3-8b"]`
     - model_max_length: Maximum sequence length. Sequences will be right padded (and
       possibly truncated)
     - number_gpus: The effective number of GPUs (to be evenly distributed to
@@ -914,6 +914,7 @@ adaptation.
     - `mistral-123b-v2`: `["q_proj", "v_proj"]`
     - `mixtral-8x7b-instruct-v0.1`: `["q_proj", "v_proj"]`
     - `granite-3-8b`: `["q_proj", "v_proj"]`
+    - `granite-3.3-8b`: `["q_proj", "v_proj"]`
     - `granite-3.1-2b`: `["q_proj", "v_proj"]`
     - `granite-3.1-8b-instruct`: `["q_proj", "v_proj"]`
     - `granite-4.0-micro`: `["q_proj", "v_proj"]`
@@ -1080,7 +1081,7 @@ memory constrained environments.
 
     - model_name: Supported models:
       <!-- markdownlint-disable-next-line line-length -->
-      `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]`
+      `["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny", "granite-3.3-8b"]`
     - model_max_length: Maximum sequence length. Sequences will be right padded (and
       possibly truncated)
     - number_gpus: The effective number of GPUs (to be evenly distributed to