metadata
tags:
- setfit
- sentence-transformers
- text-classification
- generated_from_setfit_trainer
widget:
- text: "Technology is also playing a pivotal role. GPS-enabled waste collection vehicles, mobile apps for \ncitizen grievances, and RFID-tagged bins are improving efficiency and accountability. In Surat, \nreal-time dashboards track the status of each waste collection route, helping civic staff respond to \nmissed pickups within hours. Mobile apps like â\x80\x9CSwachhta Sarathiâ\x80\x9D allow citizens to schedule bulk \nwaste pickups or report overflowing bins. \nSegregation remains a cornerstone of effective waste management. Behavior change campaigns \ninvolving school children, resident welfare associations, and self-help groups have helped increase \nawareness. In Indore and Ambikapur, waste warriors â\x80\x94 mostly women â\x80\x94 conduct daily household \nvisits, offering feedback on segregation and distributing color-coded bins. These cities now boast \n95%+ compliance in source segregation. \nPartnerships with the informal sector have also evolved. Waste pickers, who once operated without \nsafety or recognition, are being integrated into formal systems. Municipalities in Pune and Nagpur \nhave signed MoUs with waste picker cooperatives, offering ID cards, training, and inclusion in \npension and insurance schemes. This approach not only improves livelihoods but ensures better \nmaterial recovery rates. \nPlastic waste, a growing concern, is being addressed through extended producer responsibility \n(EPR) frameworks and plastic reuse innovations. Startups in Bhubaneswar and Vadodara are \nturning low-value plastic into construction material, such as plastic paver blocks and tiles. In \nUdaipur, a plastic buy-back initiative pays citizens â\x82¹5/kg for clean, dry plastic, promoting better \nhousehold-level segregation. \nBiomedical and hazardous waste management has improved through tighter regulations and \nmonitoring. Cities with medical colleges and large hospitals now operate dedicated biomedical \nwaste treatment facilities. In Varanasi, over 400 hospitals and clinics are geo-tagged and monitored \nfor waste disposal compliance, reducing risks of infection and illegal dumping. \nCommunity engagement continues to drive innovation. In Ranchi, school â\x80\x9Cgreen brigadesâ\x80\x9D conduct \nweekly clean-up drives and environmental education sessions. In Davanagere, a â\x80\x9CZero Waste Wardâ\x80\x9D \nmodel incentivizes citizens through tax rebates and public recognition for sustained segregation and \ncomposting practices. These grassroots efforts build a sense of ownership and civic responsibility. \nPublic-private partnerships are gaining traction. Many cities now engage private agencies for \noperation and maintenance of waste processing units, often under performance-based contracts. \nThese partnerships bring technical expertise and efficiency, though transparency in procurement and \nmonitoring remains crucial. \nLooking ahead, the emphasis is shifting toward circular economy principles. Instead of just \ndisposal, cities are exploring how waste can become a resource. Compost, recyclables, bio-CNG, \nand refuse-derived fuel (RDF) are being extracted from waste streams. Smart cities like Surat and \nBhopal are investing in waste-to-energy projects, aiming to reduce landfill dependency and generate \nclean energy. \nCapacity building of municipal staff is essential for sustainability. Training modules on waste \nauditing, decentralized processing, and public engagement are being rolled out through the National \nUrban Learning Platform. Cities are also setting up â\x80\x9CSolid Waste Management Cellsâ\x80\x9D within \nmunicipal corporations to ensure focused, expert-led implementation. \n \n"
- text: "Routine Operational and Supply Update â\x80\x94 Eastern Zone Sector \nDate: 25-09-2017 \nLocation: Forward Supply Base Delta-6, Eastern Command Theater \nSummary: \nThis document serves as a consolidated update on routine operations, supply chain integrity, \npersonnel morale, and minor disciplinary matters across sub-sector E1-E4 of the Eastern Theater. \nWhile it does not contain high-level intelligence, the details herein are to remain restricted to \nauthorized personnel due to potential administrative sensitivities. \nPersonnel & Discipline Update: \nâ\x80¢ Three cases of late check-ins at FOB Lambda were reported. Following standard protocol, \nthe personnel were reprimanded and assigned additional perimeter watch shifts. \nâ\x80¢ One junior officer from the logistics crew at Base Echo-7 has been reassigned after an \ninternal inquiry concluded misuse of priority requisition tags for personal gain. A follow-up \nreview is being conducted. \nâ\x80¢ Morale surveys conducted at four key outposts indicate a minor dip in satisfaction ratings \n(avg. 76.4%) due to extended deployment without rotation. Recommendations for 2-week \nstaggered rest cycles have been submitted to Command for approval. \nMeteorological Observations: \nWeather across Sector E remains stable, with isolated reports of electrical storms over Grid Point \nE17. The Meteorological Unit has issued advisories for supply convoys to reroute via E2-Highland \nbypass during storm activity. \nNotably, data from weather drones confirm a temperature anomaly over Ridge Valley â\x80\x94 a +3.7°C \nsustained temperature increase over three consecutive nights. Though not presently a tactical \nconcern, the anomaly is being monitored for environmental impact. \nSupply Chain & Resource Allocation: \nRoutine inspection of stockpiles at Base Delta-6, Gamma-2, and Omega Outpost revealed: \nâ\x80¢ Rations and medical supplies sufficient for 34 days. \nâ\x80¢ Ammunition resupply is 94% complete; minor delay in delivery of Class C explosives due \nto transportation backlog at the central depot. \nâ\x80¢ Water purification units operating at full capacity, with minor repairs completed on two \nfiltration units at Base Gamma-2. \nA new consignment of field repair kits and reinforced body armor (Batch T9-B) is en route and \nexpected within 72 hours. \n \n \n \n"
- text: "Field Surveillance System Technical Manual Review and Deployment \nBrief \nIn light of recent feedback from forward-operating units and maintenance crews, a comprehensive \nrevision of the technical manuals for the Modular Terrain Surveillance System (MTSS) has been \ncompleted. This system, deployed widely across sectors with elevated perimeter security \nrequirements, has been integral in early threat detection, terrain mapping, and signal triangulation \nefforts. However, inconsistencies in field usage, maintenance outcomes, and operator feedback \nnecessitated a structured audit and documentation overhaul. \nThe review, led by the Systems Operations Branch in conjunction with technical representatives \nfrom the original equipment manufacturer (OEM), focused on bridging operational ambiguities \nfound in the previous version of the manual, particularly in troubleshooting procedures, diagnostics \ninterpretation, and component interchangeability across variants. \nThe audit team analyzed incident logs from 27 installations over a 12-month period, identifying \npatterns in system malfunctions. The most frequent category involved erratic infrared detection \noutput, especially during rapid temperature shifts at dusk or dawn. While environmental variables \nplayed a role, improper calibration procedures were cited in over 60% of these incidents. The \nupdated manual now includes expanded calibration flowcharts, automated routine instructions, and \nconditional response matrices based on region-specific climate profiles. \nAnother prominent issue addressed is the mismatch between sensor serial batches and firmware \nupdate compatibility. In multiple cases, field teams attempted to apply a universal firmware package \nacross sensors from different production runs, causing system stalling and partial data blackouts. \nThe new documentation now includes a detailed compatibility table and introduces a secure \nverification script that flags incompatible firmware uploads before execution. \nMaintenance teams also reported inconsistent power draw behavior in the multi-sensor interface \nmodules, often leading to false fault indicators. Analysis revealed that cable shielding degradation \nin high-moisture environments caused electrical noise that mimicked hardware failure signals. The \nmanual now prescribes updated shielding inspection intervals and includes photographic \nbenchmarks of early corrosion stages to aid quicker field diagnosis. \nA key upgrade in the manual involves the layout of emergency override procedures. Previously \nfragmented across appendices, all critical fallback configurationsâ\x80\x94including manual sensor resets, \nnetwork bypass triggers, and local-only data retention switchesâ\x80\x94have been consolidated into a \nsingle quick-reference section with diagrammatic support. This change follows multiple after-action \nreports that highlighted response delays during high-tempo operations where immediate access to \noverride protocols was critical. \nThe revised manual also better accounts for the evolving deployment scenarios. With MTSS \nsystems increasingly being installed in remote and rugged terrain, new field tips have been added \nfor temporary stabilization on uneven or loosely packed soil. These include anchor plate \nschematics, modular support frames, and improvised stabilization strategies using locally available \nmaterialsâ\x80\x94all vetted through live field trials during the last joint equipment readiness event. \nOperator feedback was instrumental in refining interface terminology. Numerous field users \nreported confusion between \"signal attenuation\" and \"signal loss,\" which in prior documentation \nwere often interchanged or presented with minimal differentiation. The manual now includes \nprecise technical definitions, updated UI screenshots reflecting firmware version 6.2+, and standard \nmessaging formats for logbook entriesâ\x80\x94ensuring that incident reporting aligns accurately with \nbackend diagnostics. \n"
- text: "Overview of Regional Infrastructure Development Plans (2024â\x80\x932026) \nThe Ministry of Infrastructure and Rural Development has recently finalized the first phase of its \ncomprehensive Regional Infrastructure Development Plans for the fiscal years 2024â\x80\x932026. These \nplans emphasize a broad-based approach to rural modernization, improving logistics efficiency, and \nreinforcing national integration through enhanced connectivity. While the emphasis remains on \nboosting economic potential in tier-II and tier-III regions, urban expansion corridors will also see \ncapacity upgrades through multimodal projects. \nThe strategic planning process was initiated in Q1 2023 with a public consultation period spanning \n90 days, during which various stakeholder groups, including municipal corporations, state public \nworks departments, environmental agencies, and citizen forums, submitted their feedback. These \ninputs were analyzed and integrated into the first draft of the regional development strategy, \nfocusing on three main pillars: connectivity, sustainability, and community integration. \nOn the connectivity front, over 5,000 kilometers of feeder roads are scheduled for upgrading to all-\nweather concrete roads. Additionally, 42 key railway stations in emerging economic zones will \nundergo modernization including digital ticketing infrastructure, platform extensions, and improved \naccessibility for differently-abled citizens. Expansion plans also include the establishment of four \nnew dry ports and the refurbishment of 27 rural airstrips for short-haul cargo and emergency \nresponse functions. \nSustainability has been integrated through extensive environmental impact assessments (EIAs), \nwhich are now mandatory for every major highway and bridge project under the revised 2023 \nInfrastructure Code. EIA data will be updated and published biannually to ensure public visibility \nand cross-agency coordination. Projects situated near ecological sensitive zones, including riparian \nreserves and forest buffer belts, will be subject to stricter scrutiny under the new joint guidelines \nissued by the Ministry of Environment and the National Infrastructure Coordination Committee. \nCommunity integration has taken the form of local resource training hubs being set up in 86 rural \ndistricts. These centers are designed to upskill local populations in infrastructure maintenance, \nequipment handling, and site safety practices. With 67% of the contracted labor for the Phase-1 \nprojects expected to be sourced locally, these initiatives aim to create employment and economic \nmobility while reducing project dependency on external labor pools. \nDigital dashboards have also been introduced to improve public engagement. Citizens can now \naccess real-time updates on project progress, submit grievances or suggestions, and monitor budget \nutilization through the â\x80\x9CInfraConnectâ\x80\x9D portal. Additionally, a third-party auditing body has been \nbrought on board to verify the milestones and timelines published by the project agencies. These \ntransparency measures are expected to contribute to both time and cost efficiencies while \nreinforcing public trust. \nIn terms of budget, a total outlay of â\x82¹19,850 crores has been allocated for the 2024â\x80\x932026 period, \nsourced primarily through central funds and assisted by state contributions and international \ninfrastructure grants. So far, 67% of the planned capital has been earmarked for transport \ninfrastructure, while 23% is allocated to energy transmission corridors and the remaining 10% to \nlogistics and warehousing upgrades. \nThe Ministry is also piloting the â\x80\x9CGreen Roadsâ\x80\x9D initiative in five states, which will use recycled \nplastic and fly ash in road construction. This aligns with the National Green Commitment signed \n"
- text: "Tactical Relocation of Electronic Warfare Units in Proximity to Northern Surveillance \nCorridor \nIn response to elevated electromagnetic probing from adversarial forces near the Eastern Himalayan \nMonitoring Array, Command Group Delta has approved the phased relocation of our primary \nElectronic Warfare (EW) units to concealed forward-operation bases within Range Sector 5. The \nrelocation includes both mobile disruption teams and sensor-jamming modules, intended to fortify \nour electronic resilience while introducing unpredictable interference into known surveillance paths. \nThe decision follows a four-week intelligence assessment conducted jointly by Signals Analysis \nTask Force and Drone Recon Group 22, which identified abnormal signal clusters consistent with \nenemy pre-intrusion mapping activity. \nThe key element in this strategic move is the redeployment of the HYDRA-V EW platformâ\x80\x94a \nmodular, multi-frequency jammer equipped with auto-resonance scanning and low-power passive \ninterference arrays. These systems were previously stationed in low-activity monitoring posts near \nGrid Point S-304, but will now be repositioned along Ridge Point Vale, which offers natural \nelevation cover and reduced electromagnetic reflection due to its basaltic soil composition. The \nHYDRA units are transported via heavy-duty camouflage convoys with ECM-shielded trailers, \nescorted by Quick Response Armor units under radio blackout protocols. \nThis operation, internally designated Ghost Crown, also involves the encryption realignment of \ncomms infrastructure in the 19th Armored Brigadeâ\x80\x99s command hierarchy. Nodes at outposts Elm 1 \nthrough Elm 4 will be upgraded with directional signal filtering, allowing secure uplinks even \nduring coordinated jamming attempts. Each node will operate under the new LunaNet secure packet \nprotocol with triple-failover encryption chaining, first introduced during Operation Sky Fuse. All \noperational parameters related to these upgrades are under Secure Directive 19-Tau and are \naccessible only to personnel with clearance level Theta-5 or higher. \nVital to this relocation effort is the insertion of false telemetry signals across known observation \nvectors. Signal Echo Drones have been launched under pre-assigned trajectories to simulate \ncommunication chatter and heat signatures of standard EW truck formations. These drones are \nremotely operated from a hardened control unit beneath Fort Asmund and are programmed to \nmaintain noise profiles that mimic live systems for up to 72 hours. This deception strategy is \nintended to redirect enemy satellite recon sweeps and throw off pattern recognition algorithms used \nby opposing signal intelligence cells. \nDuring the second phase of Ghost Crown, passive reflectors will be embedded into the local terrain \nnear the new deployment corridor. These reflectors, hidden beneath natural rock formations and \nsnow cover, will scramble long-range signal capture and misrepresent node density to thermal \nsatellite passes. Initial trials conducted last month at the Markan Test Range showed a 74% \nreduction in signal triangulation success when reflectors were deployed at strategic choke points. \nThese same reflectors will also serve as secondary detonation triggers for short-range EMP mines in \nthe event of a breach. \nTo ensure continuity of mission control, Command Center Zulu has begun issuing flash-coded \ndirectives through fiber-based Q-Link pathways. All communications are routed through triple-path \nrelays to prevent isolation or disruption. Physical orders are couriered via encrypted flash nodes \ndelivered by autonomous vehicles operating under camouflage routines mapped to regional wildlife \nmovement patterns. These systems reduce heat signature visibility while maintaining GPS-denied \nnavigation efficiency. \n"
metrics:
- accuracy
pipeline_tag: text-classification
library_name: setfit
inference: true
base_model: nomic-ai/modernbert-embed-base
model-index:
- name: SetFit with nomic-ai/modernbert-embed-base
results:
- task:
type: text-classification
name: Text Classification
dataset:
name: Unknown
type: unknown
split: test
metrics:
- type: accuracy
value: 0.9663865546218487
name: Accuracy
SetFit with nomic-ai/modernbert-embed-base
This is a SetFit model that can be used for Text Classification. This SetFit model uses nomic-ai/modernbert-embed-base as the Sentence Transformer embedding model. A LogisticRegression instance is used for classification.
The model has been trained using an efficient few-shot learning technique that involves:
- Fine-tuning a Sentence Transformer with contrastive learning.
- Training a classification head with features from the fine-tuned Sentence Transformer.
Model Details
Model Description
- Model Type: SetFit
- Sentence Transformer body: nomic-ai/modernbert-embed-base
- Classification head: a LogisticRegression instance
- Maximum Sequence Length: 8192 tokens
- Number of Classes: 4 classes
Model Sources
- Repository: SetFit on GitHub
- Paper: Efficient Few-Shot Learning Without Prompts
- Blogpost: SetFit: Efficient Few-Shot Learning Without Prompts
Model Labels
| Label | Examples |
|---|---|
| 1 |
|
| 3 |
|
| 2 |
|
| 0 |
|
Evaluation
Metrics
| Label | Accuracy |
|---|---|
| all | 0.9664 |
Uses
Direct Use for Inference
First install the SetFit library:
pip install setfit
Then you can load this model and run inference.
from setfit import SetFitModel
# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("amritzeon/setfit_modernbert_finetunednew")
# Run inference
preds = model("Routine Operational and Supply Update â Eastern Zone Sector
Date: 25-09-2017
Location: Forward Supply Base Delta-6, Eastern Command Theater
Summary:
This document serves as a consolidated update on routine operations, supply chain integrity,
personnel morale, and minor disciplinary matters across sub-sector E1-E4 of the Eastern Theater.
While it does not contain high-level intelligence, the details herein are to remain restricted to
authorized personnel due to potential administrative sensitivities.
Personnel & Discipline Update:
⢠Three cases of late check-ins at FOB Lambda were reported. Following standard protocol,
the personnel were reprimanded and assigned additional perimeter watch shifts.
⢠One junior officer from the logistics crew at Base Echo-7 has been reassigned after an
internal inquiry concluded misuse of priority requisition tags for personal gain. A follow-up
review is being conducted.
⢠Morale surveys conducted at four key outposts indicate a minor dip in satisfaction ratings
(avg. 76.4%) due to extended deployment without rotation. Recommendations for 2-week
staggered rest cycles have been submitted to Command for approval.
Meteorological Observations:
Weather across Sector E remains stable, with isolated reports of electrical storms over Grid Point
E17. The Meteorological Unit has issued advisories for supply convoys to reroute via E2-Highland
bypass during storm activity.
Notably, data from weather drones confirm a temperature anomaly over Ridge Valley â a +3.7°C
sustained temperature increase over three consecutive nights. Though not presently a tactical
concern, the anomaly is being monitored for environmental impact.
Supply Chain & Resource Allocation:
Routine inspection of stockpiles at Base Delta-6, Gamma-2, and Omega Outpost revealed:
⢠Rations and medical supplies sufficient for 34 days.
⢠Ammunition resupply is 94% complete; minor delay in delivery of Class C explosives due
to transportation backlog at the central depot.
⢠Water purification units operating at full capacity, with minor repairs completed on two
filtration units at Base Gamma-2.
A new consignment of field repair kits and reinforced body armor (Batch T9-B) is en route and
expected within 72 hours.
")
Training Details
Training Set Metrics
| Training set | Min | Median | Max |
|---|---|---|---|
| Word count | 166 | 441.7377 | 559 |
| Label | Training Sample Count |
|---|---|
| 0 | 15 |
| 1 | 16 |
| 2 | 15 |
| 3 | 15 |
Training Hyperparameters
- batch_size: (8, 8)
- num_epochs: (1, 16)
- max_steps: 50
- sampling_strategy: oversampling
- body_learning_rate: (2e-05, 1e-05)
- head_learning_rate: 0.01
- loss: CosineSimilarityLoss
- distance_metric: cosine_distance
- margin: 0.25
- end_to_end: False
- use_amp: False
- warmup_proportion: 0.1
- l2_weight: 0.01
- seed: 42
- eval_max_steps: 50
- load_best_model_at_end: False
Training Results
| Epoch | Step | Training Loss | Validation Loss |
|---|---|---|---|
| 0.02 | 1 | 0.1914 | - |
| 0.2 | 10 | 0.2452 | 0.1920 |
| 0.4 | 20 | 0.1494 | 0.1849 |
| 0.6 | 30 | 0.1523 | 0.1514 |
| 0.8 | 40 | 0.1147 | 0.1255 |
| 1.0 | 50 | 0.1071 | 0.1174 |
Framework Versions
- Python: 3.11.13
- SetFit: 1.1.3
- Sentence Transformers: 4.1.0
- Transformers: 4.48.0
- PyTorch: 2.6.0+cu124
- Datasets: 4.0.0
- Tokenizers: 0.21.4
Citation
BibTeX
@article{https://doi.org/10.48550/arxiv.2209.11055,
doi = {10.48550/ARXIV.2209.11055},
url = {https://arxiv.org/abs/2209.11055},
author = {Tunstall, Lewis and Reimers, Nils and Jo, Unso Eun Seo and Bates, Luke and Korat, Daniel and Wasserblat, Moshe and Pereg, Oren},
keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {Efficient Few-Shot Learning Without Prompts},
publisher = {arXiv},
year = {2022},
copyright = {Creative Commons Attribution 4.0 International}
}