diff --git a/README.md b/README.md index e9f313c..df227f7 100644 --- a/README.md +++ b/README.md @@ -112,12 +112,14 @@ Open-Source models can be used locally via [Ollama](https://ollama.com/) or [Loc | Label | Tokens | Vision | Notes | | ----------------- | :----: | :----: | ---------------------------------------------------------------------------- | | `llama3.3` | 128 k | | Meta's new 70B model with similar performance to the Llama 3.1 405B model. | +| `llama3.2-vision` | 128 k | ✅ | Meta's image reasoning generative models in both 11B and 90B model sizes. | | `llama3.2` | 128 k | | Meta's new 1B and 3B models optimized for multilingual dialogue use cases. | | `llama3.1` | 128 k | | Meta's 8B, 70B and 405B models with overall stronger reasoning capabilities. | | `phi4` | 16 k | | Microsoft's new Phi 4 "Medium" (14B) open model with a 16k context window. | +| `phi3.5-vision` | 128 k | ✅ | Microsoft's Phi 3 and 3.5 "Mini" (3.8B) models with vision capabilities. | | `phi3.5` / `phi3` | 128 k | | Microsoft's new Phi 3.5 "Mini" (3.8B) and Phi 3 "Medium" (14B) open models. | -| `qwen2.5` | 128 k | | Alibaba Cloud's Qwen multilingual Chinese models ranging from 0.5B to 72B. | -| `gemma2` | 128 k | | Google's Gemma 2B, 9B and 27B models featuring a brand new architecture. | +| `qwen2.5` | 32 k | | Alibaba Cloud's Qwen multilingual Chinese models ranging from 0.5B to 72B. | +| `gemma2` | 8 k | | Google's Gemma 2B, 9B and 27B models featuring a brand new architecture. | | `llava` | 4 k | ✅ | Family of Large Language-and-Vision Assistant models between 7B and 34B. | ## Future Development diff --git a/src/startup.ts b/src/startup.ts index c91950c..3877724 100644 --- a/src/startup.ts +++ b/src/startup.ts @@ -258,6 +258,8 @@ export class GladdisWelcomeModal extends Modal { dropdown .addOptions({ 'llama3.3:70b': 'Llama 3.3 70B (43 Go)', + 'llama3.2-vision:11b': 'Llama 3 V 11B (8 Go)', + 'llama3.2-vision:90b': 'Llama 3 V 90B (55 Go)', 'llama3.2:1b': 'Llama 3.2 1B (1 Go)', 'llama3.2:3b': 'Llama 3.2 3B (2 Go)', 'llama3.1:8b': 'Llama 3.1 8B (5 Go)', @@ -278,6 +280,8 @@ export class GladdisWelcomeModal extends Modal { 'llava:7b': 'Llava 1.6 7B (5 Go)', 'llava:13b': 'Llava 1.6 13B (8 Go)', 'llava:34b': 'Llava 1.6 34B (20 Go)', + 'llava-phi3:3.8b': 'Llava Phi 3.8B (3 Go)', + 'llava-llama3:8b': 'Llava Llama 8B (5 Go)', }) .setValue(this.config.default_model['Ollama']) .onChange((value) => (this.config.default_model['Ollama'] = value)) @@ -334,7 +338,10 @@ export class GladdisWelcomeModal extends Modal { 'meta-llama-3.1-8b-instruct': 'Llama 3.1 8B (5 Go)', 'meta-llama-3.1-70b-instruct': 'Llama 3.1 70B (43 Go)', 'phi-4': 'Phi 4 "Medium" (9 Go)', + 'phi-3.5-vision:vllm': 'Phi 3.5 "Vision" (2 Go)', 'phi-3.5-mini-instruct': 'Phi 3.5 "Mini" (2 Go)', + 'phi-3.5-moe-instruct': 'Phi 3.5 "MoE" (24 Go)', + 'phi-3-vision:vllm': 'Phi 3 "Vision" (2 Go)', 'phi-3-medium-4k-instruct': 'Phi 3 "Medium" (8 Go)', 'qwen2.5-0.5b-instruct': 'Qwen 2.5 0.5B (0.5 Go)', 'qwen2.5-1.5b-instruct': 'Qwen 2.5 1.5B (1 Go)', @@ -346,6 +353,7 @@ export class GladdisWelcomeModal extends Modal { 'gemma-2-27b-it': 'Gemma 2 27B (16 Go)', 'llava-1.6-vicuna': 'Llava "Vicuna" (5 Go)', 'llava-1.6-mistral': 'Llava "Mistral" (5 Go)', + 'llava-llama-3-8b-v1_1': 'Llava "Llama 3" (5 Go)', }) .setValue(this.config.default_model['LocalAI']) .onChange((value) => (this.config.default_model['LocalAI'] = value)) diff --git a/src/types/catalog.ts b/src/types/catalog.ts index 44e9862..bdc4543 100644 --- a/src/types/catalog.ts +++ b/src/types/catalog.ts @@ -42,6 +42,11 @@ export const modelCatalog: ModelDetails[] = [ label: 'gpt-3.5-turbo', limit: 16_385, }, + { + label: 'llama-?3.?.?-vision', + limit: 128_000, + vision: true, + }, { label: '(meta-)?llama-?3', limit: 128_000, @@ -52,7 +57,12 @@ export const modelCatalog: ModelDetails[] = [ }, { label: 'phi-?4', - limit: 16_000, + limit: 16_384, + }, + { + label: 'phi-?3.?.?-vision', + limit: 128_000, + vision: true, }, { label: 'phi-?3', @@ -62,17 +72,9 @@ export const modelCatalog: ModelDetails[] = [ label: 'phi', limit: 2_048, }, - { - label: 'qwen2', - limit: 128_000, - }, { label: 'qwen', - limit: 8_192, - }, - { - label: 'gemma-?2', - limit: 128_000, + limit: 32_768, }, { label: 'gemma',