#lmdb — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #lmdb, aggregated by home.social.
-
Re-implementing SuperTonic for multilingual TTS - engineering notes from 5 days of training
Pretraining a 7-language flow-matching TTS (English, Vietnamese, Chinese, Korean, Japanese, German, French) on 2× RTX 4090. The training corpus: 14,633 hours of audio across 6.78M samples - roughly 15× what the original paper trained on.
Unified pipeline: parquet index + #LMDB phoneme cache.
-
Re-implementing SuperTonic for multilingual TTS - engineering notes from 5 days of training
Pretraining a 7-language flow-matching TTS (English, Vietnamese, Chinese, Korean, Japanese, German, French) on 2× RTX 4090. The training corpus: 14,633 hours of audio across 6.78M samples - roughly 15× what the original paper trained on.
Unified pipeline: parquet index + #LMDB phoneme cache.
-
Re-implementing SuperTonic for multilingual TTS - engineering notes from 5 days of training
Pretraining a 7-language flow-matching TTS (English, Vietnamese, Chinese, Korean, Japanese, German, French) on 2× RTX 4090. The training corpus: 14,633 hours of audio across 6.78M samples - roughly 15× what the original paper trained on.
Unified pipeline: parquet index + #LMDB phoneme cache.
-
Re-implementing SuperTonic for multilingual TTS - engineering notes from 5 days of training
Pretraining a 7-language flow-matching TTS (English, Vietnamese, Chinese, Korean, Japanese, German, French) on 2× RTX 4090. The training corpus: 14,633 hours of audio across 6.78M samples - roughly 15× what the original paper trained on.
Unified pipeline: parquet index + #LMDB phoneme cache.
-
Re-implementing SuperTonic for multilingual TTS - engineering notes from 5 days of training
Pretraining a 7-language flow-matching TTS (English, Vietnamese, Chinese, Korean, Japanese, German, French) on 2× RTX 4090. The training corpus: 14,633 hours of audio across 6.78M samples - roughly 15× what the original paper trained on.
Unified pipeline: parquet index + #LMDB phoneme cache.
-
More on Anvil's new datasets in HPCWire https://www.hpcwire.com/off-the-wire/purdues-anvil-streamlines-ai-research-with-ready-to-use-hpc-data-repositories/
-
More on Anvil's new datasets in HPCWire https://www.hpcwire.com/off-the-wire/purdues-anvil-streamlines-ai-research-with-ready-to-use-hpc-data-repositories/
-
More on Anvil's new datasets in HPCWire https://www.hpcwire.com/off-the-wire/purdues-anvil-streamlines-ai-research-with-ready-to-use-hpc-data-repositories/
-
More on Anvil's new datasets in HPCWire https://www.hpcwire.com/off-the-wire/purdues-anvil-streamlines-ai-research-with-ready-to-use-hpc-data-repositories/
-
More on Anvil's new datasets in HPCWire https://www.hpcwire.com/off-the-wire/purdues-anvil-streamlines-ai-research-with-ready-to-use-hpc-data-repositories/
-
Goldilocks K-Points: ML Models for Predicting K-Point Density in DFT Calculations
provides machine learning models to predict optimal k-point density (k-dist) for SCF total energy calculations with plane-wave DFT codes for inorganic 3D materials. All models take as input the structure and/or composition of the compound and output k-dist, which is expected to guarantee convergence of total energy calculations while minimizing computational time. - Built on #LMDB
https://github.com/stfc/goldilocks_kpoints -
Goldilocks K-Points: ML Models for Predicting K-Point Density in DFT Calculations
provides machine learning models to predict optimal k-point density (k-dist) for SCF total energy calculations with plane-wave DFT codes for inorganic 3D materials. All models take as input the structure and/or composition of the compound and output k-dist, which is expected to guarantee convergence of total energy calculations while minimizing computational time. - Built on #LMDB
https://github.com/stfc/goldilocks_kpoints -
Goldilocks K-Points: ML Models for Predicting K-Point Density in DFT Calculations
provides machine learning models to predict optimal k-point density (k-dist) for SCF total energy calculations with plane-wave DFT codes for inorganic 3D materials. All models take as input the structure and/or composition of the compound and output k-dist, which is expected to guarantee convergence of total energy calculations while minimizing computational time. - Built on #LMDB
https://github.com/stfc/goldilocks_kpoints -
Goldilocks K-Points: ML Models for Predicting K-Point Density in DFT Calculations
provides machine learning models to predict optimal k-point density (k-dist) for SCF total energy calculations with plane-wave DFT codes for inorganic 3D materials. All models take as input the structure and/or composition of the compound and output k-dist, which is expected to guarantee convergence of total energy calculations while minimizing computational time. - Built on #LMDB
https://github.com/stfc/goldilocks_kpoints -
Goldilocks K-Points: ML Models for Predicting K-Point Density in DFT Calculations
provides machine learning models to predict optimal k-point density (k-dist) for SCF total energy calculations with plane-wave DFT codes for inorganic 3D materials. All models take as input the structure and/or composition of the compound and output k-dist, which is expected to guarantee convergence of total energy calculations while minimizing computational time. - Built on #LMDB
https://github.com/stfc/goldilocks_kpoints -
Bharat Scene Text Dataset
We introduce the Bharat Scene Text Dataset (BSTD) — a large-scale benchmark for Indian language scene text recognition, consisting of 6,582 scene images featuring 1,26,292 words in 11 Indian languages and English. Each image is manually annotated with polygon-level bounding boxes and corresponding transcription and script, ensuring high-quality data for research and applications. - Built on #LMDB
https://github.com/Bhashini-IITJ/BharatSceneTextDataset -
Bharat Scene Text Dataset
We introduce the Bharat Scene Text Dataset (BSTD) — a large-scale benchmark for Indian language scene text recognition, consisting of 6,582 scene images featuring 1,26,292 words in 11 Indian languages and English. Each image is manually annotated with polygon-level bounding boxes and corresponding transcription and script, ensuring high-quality data for research and applications. - Built on #LMDB
https://github.com/Bhashini-IITJ/BharatSceneTextDataset -
Bharat Scene Text Dataset
We introduce the Bharat Scene Text Dataset (BSTD) — a large-scale benchmark for Indian language scene text recognition, consisting of 6,582 scene images featuring 1,26,292 words in 11 Indian languages and English. Each image is manually annotated with polygon-level bounding boxes and corresponding transcription and script, ensuring high-quality data for research and applications. - Built on #LMDB
https://github.com/Bhashini-IITJ/BharatSceneTextDataset -
Bharat Scene Text Dataset
We introduce the Bharat Scene Text Dataset (BSTD) — a large-scale benchmark for Indian language scene text recognition, consisting of 6,582 scene images featuring 1,26,292 words in 11 Indian languages and English. Each image is manually annotated with polygon-level bounding boxes and corresponding transcription and script, ensuring high-quality data for research and applications. - Built on #LMDB
https://github.com/Bhashini-IITJ/BharatSceneTextDataset -
Bharat Scene Text Dataset
We introduce the Bharat Scene Text Dataset (BSTD) — a large-scale benchmark for Indian language scene text recognition, consisting of 6,582 scene images featuring 1,26,292 words in 11 Indian languages and English. Each image is manually annotated with polygon-level bounding boxes and corresponding transcription and script, ensuring high-quality data for research and applications. - Built on #LMDB
https://github.com/Bhashini-IITJ/BharatSceneTextDataset -
Stroom is a data processing, storage and analysis platform. It is scalable - just add more CPUs / servers for greater throughput. It is suitable for processing high volume data such as system logs, to provide valuable insights into IT performance and usage.
https://github.com/gchq/stroom
Noteworthy to me because it uses #LMDB, and because it's developed by GCHQ: "We are the UK's intelligence, security and cyber agency. Our mission is to help keep the country safe."
-
Stroom is a data processing, storage and analysis platform. It is scalable - just add more CPUs / servers for greater throughput. It is suitable for processing high volume data such as system logs, to provide valuable insights into IT performance and usage.
https://github.com/gchq/stroom
Noteworthy to me because it uses #LMDB, and because it's developed by GCHQ: "We are the UK's intelligence, security and cyber agency. Our mission is to help keep the country safe."
-
Stroom is a data processing, storage and analysis platform. It is scalable - just add more CPUs / servers for greater throughput. It is suitable for processing high volume data such as system logs, to provide valuable insights into IT performance and usage.
https://github.com/gchq/stroom
Noteworthy to me because it uses #LMDB, and because it's developed by GCHQ: "We are the UK's intelligence, security and cyber agency. Our mission is to help keep the country safe."
-
Stroom is a data processing, storage and analysis platform. It is scalable - just add more CPUs / servers for greater throughput. It is suitable for processing high volume data such as system logs, to provide valuable insights into IT performance and usage.
https://github.com/gchq/stroom
Noteworthy to me because it uses #LMDB, and because it's developed by GCHQ: "We are the UK's intelligence, security and cyber agency. Our mission is to help keep the country safe."
-
Stroom is a data processing, storage and analysis platform. It is scalable - just add more CPUs / servers for greater throughput. It is suitable for processing high volume data such as system logs, to provide valuable insights into IT performance and usage.
https://github.com/gchq/stroom
Noteworthy to me because it uses #LMDB, and because it's developed by GCHQ: "We are the UK's intelligence, security and cyber agency. Our mission is to help keep the country safe."
-
Natural Language Generation from Wikidata—Architecture, Scalability and Challenges
We present an architecture for generation of Wikipedia articles in several languages from Wikidata. The articles cover the topics of countries, cities, people, universities and professions, and vary in size depending on the amount of available information. The architecture is based on the Grammatical Framework, ... Built on #LMDB
https://link.springer.com/chapter/10.1007/978-3-032-21143-9_1
-
Natural Language Generation from Wikidata—Architecture, Scalability and Challenges
We present an architecture for generation of Wikipedia articles in several languages from Wikidata. The articles cover the topics of countries, cities, people, universities and professions, and vary in size depending on the amount of available information. The architecture is based on the Grammatical Framework, ... Built on #LMDB
https://link.springer.com/chapter/10.1007/978-3-032-21143-9_1
-
Natural Language Generation from Wikidata—Architecture, Scalability and Challenges
We present an architecture for generation of Wikipedia articles in several languages from Wikidata. The articles cover the topics of countries, cities, people, universities and professions, and vary in size depending on the amount of available information. The architecture is based on the Grammatical Framework, ... Built on #LMDB
https://link.springer.com/chapter/10.1007/978-3-032-21143-9_1
-
Natural Language Generation from Wikidata—Architecture, Scalability and Challenges
We present an architecture for generation of Wikipedia articles in several languages from Wikidata. The articles cover the topics of countries, cities, people, universities and professions, and vary in size depending on the amount of available information. The architecture is based on the Grammatical Framework, ... Built on #LMDB
https://link.springer.com/chapter/10.1007/978-3-032-21143-9_1
-
Natural Language Generation from Wikidata—Architecture, Scalability and Challenges
We present an architecture for generation of Wikipedia articles in several languages from Wikidata. The articles cover the topics of countries, cities, people, universities and professions, and vary in size depending on the amount of available information. The architecture is based on the Grammatical Framework, ... Built on #LMDB
https://link.springer.com/chapter/10.1007/978-3-032-21143-9_1
-
GoldenDict-ng is an advanced dictionary lookup program, supporting many formats.
https://github.com/xiaoyifang/goldendict-ng
Uses #LMDB for indexing
-
GoldenDict-ng is an advanced dictionary lookup program, supporting many formats.
https://github.com/xiaoyifang/goldendict-ng
Uses #LMDB for indexing
-
GoldenDict-ng is an advanced dictionary lookup program, supporting many formats.
https://github.com/xiaoyifang/goldendict-ng
Uses #LMDB for indexing
-
GoldenDict-ng is an advanced dictionary lookup program, supporting many formats.
https://github.com/xiaoyifang/goldendict-ng
Uses #LMDB for indexing
-
GoldenDict-ng is an advanced dictionary lookup program, supporting many formats.
https://github.com/xiaoyifang/goldendict-ng
Uses #LMDB for indexing
-
Programmers rely on #LMDB more than redb anyway https://mastodon.social/@hyc/116231741396101301
-
Programmers rely on #LMDB more than redb anyway https://mastodon.social/@hyc/116231741396101301
-
Programmers rely on #LMDB more than redb anyway https://mastodon.social/@hyc/116231741396101301
-
Programmers rely on #LMDB more than redb anyway https://mastodon.social/@hyc/116231741396101301
-
Programmers rely on #LMDB more than redb anyway https://mastodon.social/@hyc/116231741396101301
-
Even without the vibecoding angle, redb is too limited compared to #LMDB https://mastodon.social/@hyc/116544134301056851
-
Even without the vibecoding angle, redb is too limited compared to #LMDB https://mastodon.social/@hyc/116544134301056851
-
Even without the vibecoding angle, redb is too limited compared to #LMDB https://mastodon.social/@hyc/116544134301056851
-
Even without the vibecoding angle, redb is too limited compared to #LMDB https://mastodon.social/@hyc/116544134301056851
-
Even without the vibecoding angle, redb is too limited compared to #LMDB https://mastodon.social/@hyc/116544134301056851