API ya bure ya embeddings
API ya embeddings hubadilisha maandishi kuwa vekta kwa ajili ya utafutaji, RAG na kuunganisha vitu pamoja. Watoa huduma kadhaa hutoa embeddings bila malipo. FreeLLMAPI huweka modeli 41 za bure za embeddings nyuma ya endpoint moja ya /v1/embeddings. Mwongozo huu unaonyesha modeli, vikomo vyake na msimbo unaofanya kazi.
Msimbo, majina ya modeli na lebo za katalogi ya moja kwa moja zinaonyeshwa kwa Kiingereza.
Embeddings za bure zinavyofanya kazi katika FreeLLMAPI
Endpoint ya /v1/embeddings hutumia muundo wa ombi wa OpenAI. SDK yoyote ya OpenAI inaweza kuiita. Unatuma maandishi, nawe unapata vekta moja kwa kila mfuatano wa ingizo.
Embeddings hutumia sheria tofauti ya failover kuliko gumzo. Maombi ya gumzo yanaweza kuhamia modeli nyingine. Maombi ya embeddings hayahami kamwe kwenda modeli nyingine. Vekta kutoka modeli mbili hazioani, kwa hiyo mabadiliko ya modeli kimya kimya yangeharibu hifadhi yako ya vekta.
- Familia: FreeLLMAPI hupanga watoa huduma kwa familia. Familia ni utambulisho mmoja wa modeli wenye ukubwa mmoja wa vekta, kwa mfano
bge-m3yenye vipimo 1024. - Failover: mtoa huduma mmoja katika familia anapofikia kikomo chake, kipanga njia hujaribu mtoa huduma anayefuata anayetoa familia hiyo hiyo.
- Chaguo-msingi:
model: "auto"hutumia familia ya chaguo-msingi. Unaweka familia ya chaguo-msingi kwenye ukurasa wa Models, Embeddings wa dashibodi.
Modeli za bure za embeddings leo
Jedwali linaorodhesha kila modeli ya bure ya embeddings kwenye katalogi ya FreeLLMAPI. Data inatoka kwenye katalogi hai. "Max input" ni idadi ya juu ya tokeni kwa mfuatano mmoja wa ingizo. "Free limit" ni kiwango cha bure ambacho mtoa huduma anachapisha.
| Modeli | Mtoa huduma | Vipimo | Ingizo la juu zaidi | Kikomo cha bure |
|---|---|---|---|---|
Gemini Embeddinggemini-embedding-001 | Google AI Studio | 3072 | 2048 | 100 rpm · 1K req/day |
Nemotron Embed VL 1Bnvidia/llama-nemotron-embed-vl-1b-v2 | NVIDIA NIM | 2048 | 8192 | ~40 rpm |
Nemotron 3 Embed 1B (OR free)nvidia/nemotron-3-embed-1b:free | OpenRouter | 2048 | 32768 | 20 rpm · shared free daily cap |
SEA-LION E5 Embedding 600Maisingapore/SEA-LION-E5-Embedding-600M | SEA-LION | 1024 | 8192 | free · 10 rpm |
SEA-LION ModernBERT Embedding 300Maisingapore/SEA-LION-ModernBERT-Embedding-300M | SEA-LION | 1024 | 8192 | free · 10 rpm |
SEA-LION ModernBERT Embedding 600Maisingapore/SEA-LION-ModernBERT-Embedding-600M | SEA-LION | 1024 | 8192 | free · 10 rpm |
BGE-M3@cf/baai/bge-m3 | Cloudflare Workers AI | 1024 | 8192 | 10K neurons/day (shared) |
EmbeddingGemma 300M@cf/google/embeddinggemma-300m | Cloudflare Workers AI | 768 | 2048 | 10K neurons/day (shared) |
Qwen3 Embedding 0.6B@cf/qwen/qwen3-embedding-0.6b | Cloudflare Workers AI | 1024 | 8192 | 10K neurons/day (shared) |
Cohere Embed v4embed-v4.0 | Cohere | 1536 | 128000 | 1K calls/mo (shared w/ chat) |
Gemini Embedding 2gemini-embedding-2 | Google AI Studio | 3072 | 8192 | free-tier rate limited |
Nemotron 3 Embed 1Bnvidia/nemotron-3-embed-1b | NVIDIA NIM | 2048 | 4096 | ~40 rpm |
PLaMo Embedding 1B (CF)@cf/pfnet/plamo-embedding-1b | Cloudflare Workers AI | 2048 | 4096 | 10K neurons/day (shared) |
BGE Small EN v1.5 (CF)@cf/baai/bge-small-en-v1.5 | Cloudflare Workers AI | 384 | 512 | 10K neurons/day (shared) |
BGE Base EN v1.5 (CF)@cf/baai/bge-base-en-v1.5 | Cloudflare Workers AI | 768 | 512 | 10K neurons/day (shared) |
BGE Large EN v1.5 (CF)@cf/baai/bge-large-en-v1.5 | Cloudflare Workers AI | 1024 | 512 | 10K neurons/day (shared) |
Mistral Embedmistral-embed | Mistral | 1024 | 8192 | free mode · monthly cap (shared) |
Codestral Embedcodestral-embed | Mistral | 1536 | 8192 | free mode · monthly cap (shared) |
Llama Text Embed v2 (Pinecone)llama-text-embed-v2 | Pinecone | 1024 | 2048 | 5M tokens/mo per model |
Multilingual E5 Large (Pinecone)multilingual-e5-large | Pinecone | 1024 | 507 | 5M tokens/mo per model |
Nemotron 3 Embed 1B (Speka)nvidia/nemotron-3-embed-1b | Speka | 2048 | 8192 | $1/month shared across Speka models |
OpenAI: Text Embedding 3 Small (ElectronHub)text-embedding-3-small | ElectronHub | 1536 | 8192 | $0.25/week shared free credits |
OpenAI: Text Embedding 3 Large (ElectronHub)text-embedding-3-large | ElectronHub | 3072 | 8192 | $0.25/week shared free credits |
Nemotron 3 Embed 8B (Inferbase)nvidia-nemotron-3-embed-8b | Inferbase | 4096 | 8192 | 5,000 requests/month; 100 requests and $0.50 usage/day shared |
Qwen 3 Embedding 0.6B (Inferbase)qwen3-embedding-0-6b | Inferbase | 1024 | 8192 | 5,000 requests/month; 100 requests and $0.50 usage/day shared |
Qwen 3 Embedding 4B (Inferbase)qwen3-embedding-4b | Inferbase | 2560 | 8192 | 5,000 requests/month; 100 requests and $0.50 usage/day shared |
Qwen 3 Embedding 8B (Inferbase)qwen3-embedding-8b | Inferbase | 4096 | 8192 | 5,000 requests/month; 100 requests and $0.50 usage/day shared |
LFM2.5 Embedding 350M (LLMTR)liquid/lfm-2.5-embedding-350m-free | LLMTR | 1024 | 512 | Ongoing zero-price route; provider terms apply |
BGE-M3 (SEA-LION)BAAI/bge-m3 | SEA-LION | 1024 | 8192 | free · 10 rpm |
Cohere Embed English v3embed-english-v3.0 | Cohere | 1024 | 512 | 1K calls/mo (shared w/ chat) |
Qwen3 Embedding 8B (OVH)Qwen3-Embedding-8B | OVH AI Endpoints | 4096 | 32768 | free · anonymous 2 req/min per IP |
BGE-M3 (OVH)bge-m3 | OVH AI Endpoints | 1024 | 8192 | free · anonymous 2 req/min per IP |
BGE Multilingual Gemma2 (OVH)bge-multilingual-gemma2 | OVH AI Endpoints | 3584 | 8192 | free · anonymous 2 req/min per IP |
Mistral: Codestral Embed (ElectronHub)codestral-embed | ElectronHub | 1536 | 8192 | $0.25/week shared free credits |
Mistral: Mistral Embed (ElectronHub)mistral-embed | ElectronHub | 1024 | 8192 | $0.25/week shared free credits |
Google: Gemini Embedding 2 (ElectronHub)gemini-embedding-2 | ElectronHub | 3072 | 8192 | $0.25/week shared free credits |
Google: Gemini Embedding 001 (ElectronHub)gemini-embedding-001 | ElectronHub | 3072 | 2048 | $0.25/week shared free credits |
OpenAI: Text Embedding Ada 002 (ElectronHub)text-embedding-ada-002 | ElectronHub | 1536 | 8192 | $0.25/week shared free credits |
Cohere Embed English Light v3embed-english-light-v3.0 | Cohere | 384 | 512 | 1K calls/mo (shared w/ chat) |
Cohere Embed Multilingual v3embed-multilingual-v3.0 | Cohere | 1024 | 512 | 1K calls/mo (shared w/ chat) |
Cohere Embed Multilingual Light v3embed-multilingual-light-v3.0 | Cohere | 384 | 512 | 1K calls/mo (shared w/ chat) |
Data ya moja kwa moja kutoka katalogi ya FreeLLMAPI (toleo 2026.11.19, imekaguliwa 2026-11-19). Vikomo vya bure hubadilika mara kwa mara. Katalogi hukaguliwa kila siku.
Baadhi ya watoa huduma hushiriki kiwango kimoja kati ya gumzo na embeddings. Kwa mfano, ufunguo wa majaribio wa Cohere unaruhusu miito 1,000 ya API kila mwezi kwa endpoints zote pamoja. Cloudflare Workers AI hutoa neurons 10,000 kila siku, zinazoshirikiwa na modeli zote za Workers AI kwenye akaunti.
Hatua ya 1: Sakinisha FreeLLMAPI na uongeze funguo
FreeLLMAPI ni chanzo huria. Inaendeshwa kwenye kompyuta yako kwenye Docker. Funguo zako za watoa huduma zinabaki kwenye kompyuta yako.
Sakinisha (macOS au Linux)
curl -fsSL https://freellmapi.co/install.sh | bashFungua dashibodi kwenye http://localhost:3001. Ongeza funguo za watoa huduma wako kwenye ukurasa wa Keys. Kwa embeddings, ufunguo wa Google AI Studio na ufunguo wa Cloudflare vinatosha kuanza. Nakili ufunguo wako wa umoja kutoka ukurasa huo huo.
Hatua ya 2: Ita endpoint ya embeddings
Tuma orodha ya mifuatano ili kupata vekta kadhaa kwa ombi moja. Tumia thamani ile ile ya model kwa nyaraka zako na kwa maswali yako ya utafutaji.
curl
curl http://localhost:3001/v1/embeddings \
-H "Authorization: Bearer <unified-key>" \
-H "Content-Type: application/json" \
-d '{"model": "bge-m3", "input": ["the quick brown fox", "a fast brown fox"]}'Python: weka embeddings na linganisha maandishi mawili
import math
from openai import OpenAI
client = OpenAI(base_url="http://localhost:3001/v1", api_key="<unified-key>")
resp = client.embeddings.create(
model="bge-m3", # a family name, "auto", or a provider model id
input=["How do I reset my password?", "Steps to change a forgotten password"],
)
a, b = resp.data[0].embedding, resp.data[1].embedding
print(len(resp.data), "vectors of", len(a), "dimensions")
dot = sum(x * y for x, y in zip(a, b))
norm = math.sqrt(sum(x * x for x in a)) * math.sqrt(sum(y * y for y in b))
print("cosine similarity:", round(dot / norm, 3))Sehemu ya model inakubali aina tatu: auto, jina la familia kama bge-m3, au kitambulisho cha modeli ya mtoa huduma kama @cf/baai/bge-m3. Kitambulisho cha modeli ya mtoa huduma hubadilika kuwa familia yake, kwa hiyo failover bado inabaki ndani ya familia hiyo.
Modeli ipi ya bure ya embeddings ichague
Chagua modeli mara moja kwa kila hifadhi ya vekta. Ukibadilisha modeli baadaye, lazima uweke embeddings za nyaraka zote upya.
| Mahitaji | Familia ya modeli | Sababu |
|---|---|---|
| Maandishi mafupi ya Kiingereza au ya lugha nyingi, watoa huduma wengi | bge-m3 | Vipimo 1024, tokeni 8,192 za ingizo. Watoa huduma kadhaa wa bure wanaitoa, kwa hiyo failover ina chaguo zaidi. |
| Utafutaji wa ubora wa juu kwa ufunguo wa Google | gemini-embedding-001 | Vipimo 3072. Kikomo cha ingizo ni tokeni 2,048, kwa hiyo gawanya nyaraka ndefu kuwa vipande. |
| Nyaraka ndefu katika vekta moja | embed-v4.0 (Cohere) | Dirisha kubwa la ingizo. Ufunguo wa majaribio unaruhusu miito 1,000 tu kila mwezi, unaoshirikiwa na gumzo. |
| Vekta ndogo na utafutaji wa haraka | bge-small-en-v1.5 | Vipimo 384. Kwa Kiingereza pekee, na kikomo cha ingizo ni tokeni 512. |
| Utafutaji wa msimbo | codestral-embed | Imefunzwa kwa msimbo. Inatumia kiwango cha hali ya bure ya Mistral. |
Vekta kubwa zinagharimu hifadhi zaidi na kumbukumbu zaidi kwenye hifadhidata ya vekta. Vekta ya vipimo 3072 ni mara tatu kubwa kuliko vekta ya vipimo 1024. Kwa miradi mingi ya RAG yenye vipande chini ya milioni moja, vipimo 768 hadi 1024 hutoa matokeo mazuri.
Vikomo na mitego
- Vikomo vya ingizo: maandishi yaliyozidi thamani ya "Max input" yanaweza kukatwa au kukataliwa. Gawanya nyaraka kuwa vipande vya tokeni 300 hadi 800 kabla ya kuweka embeddings.
- Familia moja kwa hifadhi: usichanganye vekta kutoka familia mbili kwenye faharasa moja. Alama za ufanano hazina maana kati ya modeli.
- Usawazishaji: Google inasema kuwa vekta za
gemini-embedding-001zenye vipimo chini ya 3072 lazima uzisawazishe mwenyewe. Sawazisha vekta zote kabla ya kutumia dot product kama alama. - Vikomo vya kasi: tabaka za bure huhesabu maombi kwa dakika au kwa siku. Tuma makundi ya mifuatano kwenye ombi moja, si ombi moja kwa kila mfuatano.
- Tabaka zisizo na jina: OVH AI Endpoints huruhusu maombi 2 kwa dakika kwa kila anwani ya IP na kwa kila modeli bila ufunguo. Itumie kama akiba, si kwa kuorodhesha kwa wingi.
Utatuzi wa matatizo
| Tatizo | Sababu | Suluhisho |
|---|---|---|
| Hitilafu 429 wakati wa kuorodhesha kwa wingi | Kikomo cha mtoa huduma kwa familia kimefikiwa | Ongeza ufunguo wa pili wa mtoa huduma kwa familia ile ile, au tuma maombi machache kwa dakika |
| Matokeo ya utafutaji si sahihi baada ya mabadiliko | Nyaraka na maswali vinatumia familia tofauti | Tumia thamani moja ya model kwa vyote viwili. Weka embeddings za nyaraka upya ukiwa umebadilisha |
| Urefu wa vekta si kama ulivyotarajia | auto hutumia familia ya chaguo-msingi | Weka jina la familia wazi, au angalia chaguo-msingi kwenye Models, Embeddings |
| Maandishi marefu yanatoa mechi duni | Maandishi ni marefu kuliko kikomo cha ingizo | Gawanya maandishi kuwa vipande vidogo zaidi |
| Hitilafu 401 | Ufunguo wa umoja si sahihi au haupo | Nakili tena ufunguo wa umoja kutoka ukurasa wa Keys |
Maswali yanayoulizwa mara kwa mara
Je, kuna API ya bure ya embeddings?
Ndiyo. Google AI Studio, Cloudflare Workers AI, Cohere, NVIDIA, Mistral, OVH na wengine hutoa modeli za bure za embeddings zenye vikomo vya kasi. FreeLLMAPI huweka 41 kati yake nyuma ya endpoint moja inayoendana na OpenAI.
Je, ninaweza kutumia OpenAI SDK kwa embeddings za bure?
Ndiyo. Weka base URL kuwa http://localhost:3001/v1 na utumie ufunguo wako wa umoja. Kisha ita client.embeddings.create kwa jina la familia kama bge-m3.
Kwa nini FreeLLMAPI haihami kwenda modeli nyingine ya embeddings inaposhindwa?
Vekta kutoka modeli tofauti hazioani. Mabadiliko ya modeli kimya kimya yangevunja utafutaji wa ufanano. Kipanga njia hubadilisha watoa huduma tu wanaotoa familia ile ile ya modeli.
Modeli ipi ya bure ya embeddings ni bora kwa RAG?
Kwa miradi mingi, bge-m3 ni chaguo-msingi zuri. Ina vipimo 1024, kikomo cha ingizo cha tokeni 8,192 na watoa huduma kadhaa wa bure. Tumia gemini-embedding-001 ikiwa unataka vipimo 3072 na una ufunguo wa Google.
Embeddings zangu zinapaswa kuwa na vipimo vingapi?
Tumia vipimo 768 hadi 1024 kwa kazi nyingi za utafutaji na RAG. Vekta kubwa hutumia hifadhi na kumbukumbu zaidi. Chagua ukubwa mara moja, kwa sababu mabadiliko yanamaanisha lazima uweke embeddings za nyaraka zote upya.
Miongozo inayohusiana
- API ya LLM bure kwa RAG
- API ya LLM ya bure kwa LangChain
- LLM API ya bure kwa Python
- free cohere api
- Mipaka ya kiwango cha API ya LLM bure
Vyanzo
- FreeLLMAPI docs: Embeddings (imekaguliwa 20 November 2026)
- Google AI Studio docs: Embeddings (imekaguliwa 20 November 2026)
- Cloudflare Workers AI pricing (imekaguliwa 20 November 2026)
- Cohere docs: API keys and rate limits (imekaguliwa 20 November 2026)
- OVHcloud AI Endpoints capabilities (imekaguliwa 20 November 2026)