GritLM/GritLM-7B
Primitive: /encode · Encode ·
Mistral
> GritLM is a generative representational instruction tuned language model. It unifies text representation (embedding) and text generation into a single model achieving state-of-the-art performance on both types of tasks.
View on Hugging Face → Fine-tuned from mistralai/Mistral-7B-v0.1
Overview
Hardware: — drives latency, throughput & cost
| Size | 7.2B params |
|---|---|
| Tasks | /encode |
| License | apache-2.0 |
| Latency | 2.1 s |
| Throughput | 1.4K tok/s |
| Cost | $0.157 /1M tok |
Cost is approximate — computed from list GPU prices; your actual price depends on the provider you deploy SIE with.
Embedding
| Output types | Dense |
|---|---|
| Dimensions | dense: 4,096 |
| Max sequence length | 4,096 |
| Inputs | text |
Benchmarks
CQADupstackPhysicsRetrieval
Duplicate question retrieval from StackExchange Physics
CosQA
Code search with natural language queries
FiQA2018
Financial opinion mining and question answering
LegalBenchConsumerContractsQA
Question answering on consumer contracts
NFCorpus
Biomedical literature search from NutritionFacts.org
NanoFiQA2018Retrieval
Smaller subset of the FiQA financial QA dataset
SCIDOCS
Citation prediction, document classification, and recommendation for scientific papers
SciFact
Scientific claim verification using research literature