@@ -105,11 +105,102 @@ Served models can then be benchmarked using [GuideLLM](https://github.com/vllm-p
105
105
</picture >
106
106
</p >
107
107
108
- ### License
108
+ ### Supported Models
109
+
110
+ The following models are currently supported or are planned to be supported in the short term.
111
+
112
+ <table >
113
+ <thead >
114
+ <tr >
115
+ <th >Verifier Architecture</th >
116
+ <th >Verifier Size</th >
117
+ <th >Training via Speculators</th >
118
+ <th >Deployment in vLLM</th >
119
+ <th >Conversion of External Checkpoints</th >
120
+ </tr >
121
+ </thead >
122
+ <tbody >
123
+ <tr >
124
+ <td rowspan =" 3 " >Llama</td >
125
+ <td >8B-Instruct</td >
126
+ <td >EAGLE-3 ✅ | HASS ✅</td >
127
+ <td >✅</td >
128
+ <td ><a href =" https://huggingface.co/yuhuili/EAGLE3-LLaMA3.1-Instruct-8B " >EAGLE-3</a > ✅</td >
129
+ </tr >
130
+ <tr >
131
+ <td >70B-Instruct</td >
132
+ <td >EAGLE-3 ⏳</td >
133
+ <td >✅</td >
134
+ <td ><a href =" https://huggingface.co/yuhuili/EAGLE3-LLaMA3.3-Instruct-70B " >EAGLE-3</a > ✅</td >
135
+ </tr >
136
+ <tr >
137
+ <td >DeepSeek-R1-Distill-LLama-8B</td >
138
+ <td >EAGLE-3 ❌</td >
139
+ <td >✅</td >
140
+ <td ><a href =" https://huggingface.co/yuhuili/EAGLE3-DeepSeek-R1-Distill-LLaMA-8B " >EAGLE-3</a > ✅</td >
141
+ </tr >
142
+ <tr >
143
+ <td rowspan =" 3 " >Qwen3</td >
144
+ <td >8B</td >
145
+ <td >EAGLE-3 ✅</td >
146
+ <td >✅</td >
147
+ <td >❌</td >
148
+ </tr >
149
+ <tr >
150
+ <td >14B</td >
151
+ <td >EAGLE-3 ❌</td >
152
+ <td >✅</td >
153
+ <td >❌</td >
154
+ </tr >
155
+ <tr >
156
+ <td >32B</td >
157
+ <td >EAGLE-3 ❌</td >
158
+ <td >✅</td >
159
+ <td >❌</td >
160
+ </tr >
161
+ <tr >
162
+ <td rowspan =" 2 " >Qwen3 MoE</td >
163
+ <td >30B-A3B</td >
164
+ <td >EAGLE-3 ❌</td >
165
+ <td >⏳</td >
166
+ <td >❌</td >
167
+ </tr >
168
+ <tr >
169
+ <td >235B-A22B</td >
170
+ <td >EAGLE-3 ❌</td >
171
+ <td >⏳</td >
172
+ <td ><a href =" https://huggingface.co/nvidia/Qwen3-235B-A22B-Eagle3 " >EAGLE-3</a > ⏳</td >
173
+ </tr >
174
+ <tr >
175
+ <td rowspan =" 2 " >Llama-4</td >
176
+ <td >Scout-17B-16E-Instruct</td >
177
+ <td >EAGLE-3 ❌</td >
178
+ <td >⏳</td >
179
+ <td >❌</td >
180
+ </tr >
181
+ <tr >
182
+ <td >Maverick-17B-128E-Eagle3</td >
183
+ <td >EAGLE-3 ❌</td >
184
+ <td >⏳</td >
185
+ <td ><a href =" https://huggingface.co/nvidia/Llama-4-Maverick-17B-128E-Eagle3 " >EAGLE-3</a > ⏳</td >
186
+ </tr >
187
+ <tr >
188
+ <td >DeepSeek-R1</td >
189
+ <td >DeepSeek-R1</td >
190
+ <td >EAGLE-3 ❌</td >
191
+ <td >⏳</td >
192
+ <td ><a href =" https://huggingface.co/HArmonizedSS/HASS-DeepSeek-R1 " >HASS</a > ⏳</td >
193
+ </tr >
194
+ </tbody >
195
+ </table >
196
+
197
+ ✅ = Supported, ⏳ = In Progress, ❌ = Not Yet Supported
198
+
199
+ ## License
109
200
110
201
Speculators is licensed under the [ Apache License 2.0] ( https://github.com/neuralmagic/speculators/blob/main/LICENSE ) .
111
202
112
- ### Cite
203
+ ## Cite
113
204
114
205
If you find Speculators helpful in your research or projects, please consider citing it:
115
206
0 commit comments