Minor tweaks to newer models (#50)

* add deepseek-ai path, fix Smol  gsm8k value

* Qwen3 model on quad, deepseek limit gpu memory

* try Qwen3 gpu_memory_utilization

* try max-model-len

Author: derekk-nm

HuggingFaceTB/SmolLM3-3B/accuracy/tasks.yml

@@ -2,4 +2,4 @@ tasks:
   - name: gsm8k
     metrics:
       - name: exact_match,strict-match
-        value: 0
+        value: 0.4708

Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8/accuracy/server.yml

@@ -0,0 +1,3 @@
+trust-remote-code: true
+tensor-parallel-size: 4
+max-model-len: 4096

RedHatAI/DeepSeek-R1-0528-quantized.w4a16/accuracy/server.yml

@@ -1,3 +1,4 @@
 max-model-len: 4096
 tensor-parallel-size: 8
 trust-remote-code: true
+gpu_memory_utilization: 0.8

deepseek-ai/DeepSeek-R1-0528/accuracy/tasks.yml

@@ -0,0 +1,5 @@
+tasks:
+  - name: gsm8k
+    metrics:
+      - name: exact_match,strict-match
+        value: 0

deepseek-ai/DeepSeek-R1-0528/storage.yml

@@ -0,0 +1,3 @@
+# storage configs for https://huggingface.co/deepseek-ai/DeepSeek-R1-0528
+model: hf
+data: hf