Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Thomas G. Lopes
commited on
Commit
·
4148351
1
Parent(s):
a76eef5
update ctx length
Browse files- src/lib/data/context_length.json +13 -12
src/lib/data/context_length.json
CHANGED
|
@@ -15,7 +15,8 @@
|
|
| 15 |
"Meta-Llama-Guard-3-8B": 16384,
|
| 16 |
"QwQ-32B": 16384,
|
| 17 |
"Qwen2-Audio-7B-Instruct": 4096,
|
| 18 |
-
"Qwen3-32B": 8192
|
|
|
|
| 19 |
},
|
| 20 |
"nebius": {
|
| 21 |
"meta-llama/Meta-Llama-3.1-8B-Instruct-fast": 131072,
|
|
@@ -121,7 +122,7 @@
|
|
| 121 |
"deepseek/deepseek-r1-distill-qwen-14b": 64000,
|
| 122 |
"meta-llama/llama-3.3-70b-instruct": 131072,
|
| 123 |
"qwen/qwen-2.5-72b-instruct": 32000,
|
| 124 |
-
"mistralai/mistral-nemo":
|
| 125 |
"deepseek/deepseek-r1-distill-qwen-32b": 64000,
|
| 126 |
"meta-llama/llama-3-8b-instruct": 8192,
|
| 127 |
"microsoft/wizardlm-2-8x22b": 65535,
|
|
@@ -136,6 +137,7 @@
|
|
| 136 |
"cognitivecomputations/dolphin-mixtral-8x22b": 16000,
|
| 137 |
"sophosympatheia/midnight-rose-70b": 4096,
|
| 138 |
"sao10k/l3-8b-lunaris": 8192,
|
|
|
|
| 139 |
"qwen/qwen3-0.6b-fp8": 32000,
|
| 140 |
"qwen/qwen3-1.7b-fp8": 32000,
|
| 141 |
"qwen/qwen3-8b-fp8": 128000,
|
|
@@ -208,23 +210,28 @@
|
|
| 208 |
"command": 4096
|
| 209 |
},
|
| 210 |
"together": {
|
| 211 |
-
"Qwen/QwQ-32B": 131072,
|
| 212 |
"meta-llama/Llama-4-Scout-17B-16E-Instruct": 1048576,
|
| 213 |
-
"mistralai/Mistral-7B-Instruct-v0.1": 32768,
|
| 214 |
"meta-llama/Llama-Guard-4-12B": 1048576,
|
| 215 |
"togethercomputer/m2-bert-80M-32k-retrieval": 32768,
|
| 216 |
"cartesia/sonic": 0,
|
| 217 |
"scb10x/scb10x-llama3-1-typhoon2-8b-instruct": 8192,
|
| 218 |
-
"Qwen/Qwen2.5-7B-Instruct-Turbo": 32768,
|
| 219 |
"deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free": 8192,
|
| 220 |
"meta-llama-llama-2-70b-hf": 4096,
|
| 221 |
"intfloat/multilingual-e5-large-instruct": 514,
|
| 222 |
"BAAI/bge-base-en-v1.5": 512,
|
|
|
|
| 223 |
"Gryphe/MythoMax-L2-13b": 4096,
|
| 224 |
"Alibaba-NLP/gte-modernbert-base": 8192,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 225 |
"mistralai/Mixtral-8x7B-Instruct-v0.1": 32768,
|
| 226 |
"google/gemma-2-27b-it": 8192,
|
| 227 |
"Qwen/Qwen2-VL-72B-Instruct": 32768,
|
|
|
|
| 228 |
"meta-llama/LlamaGuard-2-8b": 8192,
|
| 229 |
"cartesia/sonic-2": 0,
|
| 230 |
"togethercomputer/m2-bert-80M-8k-retrieval": 8192,
|
|
@@ -240,7 +247,6 @@
|
|
| 240 |
"meta-llama/Meta-Llama-Guard-3-8B": 8192,
|
| 241 |
"marin-community/marin-8b-instruct": 131072,
|
| 242 |
"deepseek-ai/DeepSeek-R1": 163840,
|
| 243 |
-
"meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo": 131072,
|
| 244 |
"Qwen/Qwen2.5-VL-72B-Instruct": 32768,
|
| 245 |
"arcee-ai/arcee-blitz": 32768,
|
| 246 |
"arcee-ai/caller": 32768,
|
|
@@ -253,7 +259,6 @@
|
|
| 253 |
"mistralai/Mistral-Small-24B-Instruct-2501": 32768,
|
| 254 |
"meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8": 1048576,
|
| 255 |
"togethercomputer/MoA-1-Turbo": 32768,
|
| 256 |
-
"meta-llama/Llama-3.3-70B-Instruct-Turbo": 131072,
|
| 257 |
"perplexity-ai/r1-1776": 163840,
|
| 258 |
"NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO": 32768,
|
| 259 |
"deepseek-ai/DeepSeek-R1-Distill-Llama-70B": 131072,
|
|
@@ -261,23 +266,19 @@
|
|
| 261 |
"meta-llama/Meta-Llama-3-8B-Instruct-Lite": 8192,
|
| 262 |
"Qwen/Qwen3-235B-A22B-fp8": 40960,
|
| 263 |
"Qwen/Qwen3-235B-A22B-fp8-tput": 40960,
|
|
|
|
| 264 |
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": 32768,
|
| 265 |
-
"arcee_ai/arcee-spotlight": 131072,
|
| 266 |
-
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": 130815,
|
| 267 |
"scb10x/scb10x-llama3-1-typhoon2-70b-instruct": 8192,
|
| 268 |
-
"meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo": 131072,
|
| 269 |
"mistralai/Mistral-7B-Instruct-v0.2": 32768,
|
| 270 |
"deepseek-ai/DeepSeek-V3-p-dp": 131072,
|
| 271 |
"deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": 131072,
|
| 272 |
"Qwen/Qwen2.5-Coder-32B-Instruct": 16384,
|
| 273 |
"Qwen/Qwen2-72B-Instruct": 32768,
|
| 274 |
-
"meta-llama/Llama-3-70b-chat-hf": 8192,
|
| 275 |
"mistralai/Mistral-7B-Instruct-v0.3": 32768,
|
| 276 |
"Salesforce/Llama-Rank-V1": 8192,
|
| 277 |
"meta-llama/Llama-Vision-Free": 131072,
|
| 278 |
"meta-llama/Llama-Guard-3-11B-Vision-Turbo": 131072,
|
| 279 |
"meta-llama/Llama-3.2-3B-Instruct-Turbo": 131072,
|
| 280 |
-
"Qwen/Qwen2.5-72B-Instruct-Turbo": 131072,
|
| 281 |
"meta-llama/Llama-2-70b-hf": 4096
|
| 282 |
},
|
| 283 |
"fireworks-ai": {
|
|
|
|
| 15 |
"Meta-Llama-Guard-3-8B": 16384,
|
| 16 |
"QwQ-32B": 16384,
|
| 17 |
"Qwen2-Audio-7B-Instruct": 4096,
|
| 18 |
+
"Qwen3-32B": 8192,
|
| 19 |
+
"Whisper-Large-v3": 4096
|
| 20 |
},
|
| 21 |
"nebius": {
|
| 22 |
"meta-llama/Meta-Llama-3.1-8B-Instruct-fast": 131072,
|
|
|
|
| 122 |
"deepseek/deepseek-r1-distill-qwen-14b": 64000,
|
| 123 |
"meta-llama/llama-3.3-70b-instruct": 131072,
|
| 124 |
"qwen/qwen-2.5-72b-instruct": 32000,
|
| 125 |
+
"mistralai/mistral-nemo": 64000,
|
| 126 |
"deepseek/deepseek-r1-distill-qwen-32b": 64000,
|
| 127 |
"meta-llama/llama-3-8b-instruct": 8192,
|
| 128 |
"microsoft/wizardlm-2-8x22b": 65535,
|
|
|
|
| 137 |
"cognitivecomputations/dolphin-mixtral-8x22b": 16000,
|
| 138 |
"sophosympatheia/midnight-rose-70b": 4096,
|
| 139 |
"sao10k/l3-8b-lunaris": 8192,
|
| 140 |
+
"pa/cd-op-4-20250514": 200000,
|
| 141 |
"qwen/qwen3-0.6b-fp8": 32000,
|
| 142 |
"qwen/qwen3-1.7b-fp8": 32000,
|
| 143 |
"qwen/qwen3-8b-fp8": 128000,
|
|
|
|
| 210 |
"command": 4096
|
| 211 |
},
|
| 212 |
"together": {
|
|
|
|
| 213 |
"meta-llama/Llama-4-Scout-17B-16E-Instruct": 1048576,
|
|
|
|
| 214 |
"meta-llama/Llama-Guard-4-12B": 1048576,
|
| 215 |
"togethercomputer/m2-bert-80M-32k-retrieval": 32768,
|
| 216 |
"cartesia/sonic": 0,
|
| 217 |
"scb10x/scb10x-llama3-1-typhoon2-8b-instruct": 8192,
|
|
|
|
| 218 |
"deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free": 8192,
|
| 219 |
"meta-llama-llama-2-70b-hf": 4096,
|
| 220 |
"intfloat/multilingual-e5-large-instruct": 514,
|
| 221 |
"BAAI/bge-base-en-v1.5": 512,
|
| 222 |
+
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": 130815,
|
| 223 |
"Gryphe/MythoMax-L2-13b": 4096,
|
| 224 |
"Alibaba-NLP/gte-modernbert-base": 8192,
|
| 225 |
+
"mistralai/Mistral-7B-Instruct-v0.1": 32768,
|
| 226 |
+
"Qwen/Qwen2.5-7B-Instruct-Turbo": 32768,
|
| 227 |
+
"Qwen/Qwen2.5-72B-Instruct-Turbo": 131072,
|
| 228 |
+
"meta-llama/Llama-3.3-70B-Instruct-Turbo": 131072,
|
| 229 |
+
"Qwen/QwQ-32B": 131072,
|
| 230 |
+
"meta-llama/Llama-3-70b-chat-hf": 8192,
|
| 231 |
"mistralai/Mixtral-8x7B-Instruct-v0.1": 32768,
|
| 232 |
"google/gemma-2-27b-it": 8192,
|
| 233 |
"Qwen/Qwen2-VL-72B-Instruct": 32768,
|
| 234 |
+
"meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo": 131072,
|
| 235 |
"meta-llama/LlamaGuard-2-8b": 8192,
|
| 236 |
"cartesia/sonic-2": 0,
|
| 237 |
"togethercomputer/m2-bert-80M-8k-retrieval": 8192,
|
|
|
|
| 247 |
"meta-llama/Meta-Llama-Guard-3-8B": 8192,
|
| 248 |
"marin-community/marin-8b-instruct": 131072,
|
| 249 |
"deepseek-ai/DeepSeek-R1": 163840,
|
|
|
|
| 250 |
"Qwen/Qwen2.5-VL-72B-Instruct": 32768,
|
| 251 |
"arcee-ai/arcee-blitz": 32768,
|
| 252 |
"arcee-ai/caller": 32768,
|
|
|
|
| 259 |
"mistralai/Mistral-Small-24B-Instruct-2501": 32768,
|
| 260 |
"meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8": 1048576,
|
| 261 |
"togethercomputer/MoA-1-Turbo": 32768,
|
|
|
|
| 262 |
"perplexity-ai/r1-1776": 163840,
|
| 263 |
"NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO": 32768,
|
| 264 |
"deepseek-ai/DeepSeek-R1-Distill-Llama-70B": 131072,
|
|
|
|
| 266 |
"meta-llama/Meta-Llama-3-8B-Instruct-Lite": 8192,
|
| 267 |
"Qwen/Qwen3-235B-A22B-fp8": 40960,
|
| 268 |
"Qwen/Qwen3-235B-A22B-fp8-tput": 40960,
|
| 269 |
+
"meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo": 131072,
|
| 270 |
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": 32768,
|
|
|
|
|
|
|
| 271 |
"scb10x/scb10x-llama3-1-typhoon2-70b-instruct": 8192,
|
|
|
|
| 272 |
"mistralai/Mistral-7B-Instruct-v0.2": 32768,
|
| 273 |
"deepseek-ai/DeepSeek-V3-p-dp": 131072,
|
| 274 |
"deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": 131072,
|
| 275 |
"Qwen/Qwen2.5-Coder-32B-Instruct": 16384,
|
| 276 |
"Qwen/Qwen2-72B-Instruct": 32768,
|
|
|
|
| 277 |
"mistralai/Mistral-7B-Instruct-v0.3": 32768,
|
| 278 |
"Salesforce/Llama-Rank-V1": 8192,
|
| 279 |
"meta-llama/Llama-Vision-Free": 131072,
|
| 280 |
"meta-llama/Llama-Guard-3-11B-Vision-Turbo": 131072,
|
| 281 |
"meta-llama/Llama-3.2-3B-Instruct-Turbo": 131072,
|
|
|
|
| 282 |
"meta-llama/Llama-2-70b-hf": 4096
|
| 283 |
},
|
| 284 |
"fireworks-ai": {
|