ctrl g shortcut

src/lib/components/inference-playground/message.svelte

@@ -16,6 +16,7 @@
 	import LocalToasts from "../local-toasts.svelte";
 	import ImgPreview from "./img-preview.svelte";
 	import { TEST_IDS } from "$lib/constants.js";
+	import { cmdOrCtrl } from "$lib/utils/platform.js";
 
 	type Props = {
 		conversation: ConversationClass;
@@ -103,6 +104,13 @@
 					if (!message || !content) return;
 					conversation.updateMessage({ index, message: { ...message, content } });
 				}}
+				onkeydown={e => {
+					if ((e.ctrlKey || e.metaKey) && e.key === "g") {
+						e.preventDefault();
+						e.stopPropagation();
+						onRegen?.();
+					}
+				}}
 				placeholder="Enter {message?.role} message"
 				class="grow resize-none overflow-hidden rounded-lg bg-transparent px-2 py-2.5 ring-gray-100 outline-none group-hover/message:ring-3 hover:bg-white focus:bg-white focus:ring-3 @2xl:px-3 dark:ring-gray-600 dark:hover:bg-gray-900 dark:focus:bg-gray-900"
 				rows="1"
@@ -182,7 +190,15 @@
 								<IconCustom icon={message.role === "user" ? "regen" : "refresh"} />
 							</button>
 						{/snippet}
-						{regenLabel}
+						<div class="flex items-center gap-2">
+							{regenLabel}
+
+							<span
+								class="inline-flex items-center gap-0.5 rounded-sm border border-white/20 bg-white/10 px-0.5 text-xs text-white/70"
+							>
+								{cmdOrCtrl}<span class="">G</span>
+							</span>
+						</div>
 					</Tooltip>
 
 					<Tooltip>

src/lib/data/context_length.json

@@ -21,36 +21,15 @@
 	"nebius": {
 		"meta-llama/Meta-Llama-3.1-8B-Instruct-fast": 131072,
 		"meta-llama/Meta-Llama-3.1-8B-Instruct": 131072,
-		"meta-llama/Meta-Llama-3.1-70B-Instruct-fast": 131072,
 		"meta-llama/Meta-Llama-3.1-70B-Instruct": 131072,
 		"meta-llama/Meta-Llama-3.1-405B-Instruct": 131072,
 		"meta-llama/Llama-Guard-3-8B": 131072,
-		"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF-fast": 131072,
-		"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": 131072,
 		"nvidia/Llama-3_1-Nemotron-Ultra-253B-v1": 131072,
-		"mistralai/Mistral-Nemo-Instruct-2407-fast": 128000,
 		"mistralai/Mistral-Nemo-Instruct-2407": 128000,
-		"mistralai/Mixtral-8x7B-Instruct-v0.1-fast": 32768,
-		"mistralai/Mixtral-8x7B-Instruct-v0.1": 32768,
-		"mistralai/Mixtral-8x22B-Instruct-v0.1-fast": 65536,
-		"mistralai/Mixtral-8x22B-Instruct-v0.1": 65536,
-		"allenai/OLMo-7B-Instruct-hf": 2048,
-		"microsoft/Phi-3-mini-4k-instruct-fast": 4096,
-		"microsoft/Phi-3-mini-4k-instruct": 4096,
-		"microsoft/Phi-3-medium-128k-instruct-fast": 131072,
-		"microsoft/Phi-3-medium-128k-instruct": 131072,
-		"google/gemma-2-2b-it-fast": 8192,
 		"google/gemma-2-2b-it": 8192,
 		"google/gemma-2-9b-it-fast": 8192,
-		"google/gemma-2-9b-it": 8192,
-		"google/gemma-2-27b-it-fast": 8192,
-		"google/gemma-2-27b-it": 8192,
-		"deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct-fast": 128000,
-		"deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct": 128000,
 		"Qwen/Qwen2.5-Coder-7B-fast": 32768,
 		"Qwen/Qwen2.5-Coder-7B": 32768,
-		"Qwen/Qwen2.5-Coder-7B-Instruct-fast": 32768,
-		"Qwen/Qwen2.5-Coder-7B-Instruct": 32768,
 		"Qwen/Qwen2.5-Coder-32B-Instruct-fast": 131072,
 		"Qwen/Qwen2.5-Coder-32B-Instruct": 131072,
 		"Qwen/Qwen2.5-32B-Instruct-fast": 131072,
@@ -58,24 +37,12 @@
 		"Qwen/Qwen2.5-72B-Instruct-fast": 131072,
 		"Qwen/Qwen2.5-72B-Instruct": 131072,
 		"Qwen/Qwen2-VL-72B-Instruct": 32768,
-		"Qwen/Qwen2-VL-7B-Instruct": 32768,
-		"llava-hf/llava-1.5-7b-hf": 4096,
-		"llava-hf/llava-1.5-13b-hf": 4096,
-		"aaditya/Llama3-OpenBioLLM-8B": 8192,
 		"aaditya/Llama3-OpenBioLLM-70B": 8192,
 		"BAAI/bge-en-icl": 32768,
 		"BAAI/bge-multilingual-gemma2": 8192,
 		"intfloat/e5-mistral-7b-instruct": 32768,
-		"cognitivecomputations/dolphin-2.9.2-mixtral-8x22b": 65536,
-		"microsoft/Phi-3.5-MoE-instruct": 131072,
-		"microsoft/Phi-3.5-mini-instruct": 131072,
-		"Qwen/Qwen2.5-1.5B-Instruct": 32768,
 		"meta-llama/Llama-3.3-70B-Instruct": 131072,
 		"meta-llama/Llama-3.3-70B-Instruct-fast": 131072,
-		"meta-llama/Llama-3.2-1B-Instruct": 131072,
-		"meta-llama/Llama-3.2-3B-Instruct": 131072,
-		"Qwen/QwQ-32B-Preview": 32768,
-		"Qwen/QVQ-72B-preview": 128000,
 		"microsoft/phi-4": 16384,
 		"deepseek-ai/DeepSeek-V3": 163840,
 		"deepseek-ai/DeepSeek-R1": 163840,
@@ -88,6 +55,7 @@
 		"Qwen/Qwen3-30B-A3B": 40960,
 		"Qwen/Qwen3-30B-A3B-fast": 40960,
 		"Qwen/Qwen3-32B": 40960,
+		"Qwen/Qwen3-32B-fast": 40960,
 		"Qwen/Qwen3-14B": 40960,
 		"Qwen/Qwen3-4B-fast": 40960,
 		"nvidia/Llama-3_3-Nemotron-Super-49B-v1": 131072,
@@ -117,7 +85,6 @@
 		"deepseek/deepseek-prover-v2-671b": 160000,
 		"meta-llama/llama-4-scout-17b-16e-instruct": 131072,
 		"deepseek/deepseek-r1-distill-llama-8b": 32000,
-		"deepseek/deepseek_v3": 64000,
 		"meta-llama/llama-3.1-8b-instruct": 16384,
 		"deepseek/deepseek-r1-distill-qwen-14b": 64000,
 		"meta-llama/llama-3.3-70b-instruct": 131072,
@@ -131,7 +98,6 @@
 		"google/gemma-2-9b-it": 8192,
 		"mistralai/mistral-7b-instruct": 32768,
 		"meta-llama/llama-3-70b-instruct": 8192,
-		"deepseek/deepseek-r1": 64000,
 		"nousresearch/hermes-2-pro-llama-3-8b": 8192,
 		"sao10k/l3-70b-euryale-v2.1": 8192,
 		"cognitivecomputations/dolphin-mixtral-8x22b": 16000,
@@ -210,13 +176,12 @@
 		"command": 4096
 	},
 	"together": {
-		"meta-llama/Llama-4-Scout-17B-16E-Instruct": 1048576,
 		"meta-llama/Llama-Guard-4-12B": 1048576,
 		"togethercomputer/m2-bert-80M-32k-retrieval": 32768,
 		"cartesia/sonic": 0,
 		"scb10x/scb10x-llama3-1-typhoon2-8b-instruct": 8192,
-		"deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free": 8192,
 		"meta-llama-llama-2-70b-hf": 4096,
+		"togethercomputer/Refuel-Llm-V2": 16384,
 		"intfloat/multilingual-e5-large-instruct": 514,
 		"BAAI/bge-base-en-v1.5": 512,
 		"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": 130815,
@@ -233,8 +198,11 @@
 		"meta-llama/LlamaGuard-2-8b": 8192,
 		"cartesia/sonic-2": 0,
 		"togethercomputer/m2-bert-80M-8k-retrieval": 8192,
+		"deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": 131072,
+		"deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": 131072,
+		"deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free": 8192,
 		"meta-llama/Llama-3.3-70B-Instruct-Turbo-Free": 131072,
-		"togethercomputer/Refuel-Llm-V2": 16384,
+		"deepseek-ai/DeepSeek-R1-Distill-Llama-70B": 131072,
 		"deepseek-ai/DeepSeek-V3": 131072,
 		"togethercomputer/Refuel-Llm-V2-Small": 8192,
 		"togethercomputer/MoA-1": 32768,
@@ -260,8 +228,6 @@
 		"togethercomputer/MoA-1-Turbo": 32768,
 		"perplexity-ai/r1-1776": 163840,
 		"NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO": 32768,
-		"deepseek-ai/DeepSeek-R1-Distill-Llama-70B": 131072,
-		"deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": 131072,
 		"meta-llama/Meta-Llama-3-8B-Instruct-Lite": 8192,
 		"Qwen/Qwen3-235B-A22B-fp8": 40960,
 		"Qwen/Qwen3-235B-A22B-fp8-tput": 40960,
@@ -270,7 +236,6 @@
 		"scb10x/scb10x-llama3-1-typhoon2-70b-instruct": 8192,
 		"mistralai/Mistral-7B-Instruct-v0.2": 32768,
 		"deepseek-ai/DeepSeek-V3-p-dp": 131072,
-		"deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": 131072,
 		"Qwen/Qwen2.5-Coder-32B-Instruct": 16384,
 		"Qwen/Qwen2-72B-Instruct": 32768,
 		"mistralai/Mistral-7B-Instruct-v0.3": 32768,
@@ -278,9 +243,10 @@
 		"meta-llama/Llama-Vision-Free": 131072,
 		"meta-llama/Llama-Guard-3-11B-Vision-Turbo": 131072,
 		"meta-llama/Llama-3.2-3B-Instruct-Turbo": 131072,
-		"arcee_ai/arcee-spotlight": 131072,
 		"meta-llama/Llama-3.3-70B-Instruct-Turbo": 131072,
-		"meta-llama/Llama-2-70b-hf": 4096
+		"meta-llama/Llama-4-Scout-17B-16E-Instruct": 1048576,
+		"meta-llama/Llama-2-70b-hf": 4096,
+		"arcee_ai/arcee-spotlight": 131072
 	},
 	"fireworks-ai": {
 		"accounts/fireworks/models/qwq-32b": 131072,