parallax-8b-lambda

Parallax 8B lambda: mid-training checkpoints

These are research checkpoints from lambda, a training run of Parallax that is still in progress. Parallax is a mixture-of-experts language model trained on consumer GPUs spread over several countries. The hosts have no direct connections to each other and synchronize over the public internet.

• Base model. It is pretrained on a mixture of web, document, math, code, knowledge-focused and book text. It is not instruction-tuned, chat-tuned or safety-tuned, and it will continue text rather than follow instructions.
• Mid-training. Each export is a snapshot of a run that has not finished. Later exports are expected to differ, and quality between exports is not monotonic.
• Research artifact. It is published so the training can be followed and inspected. It is not intended for production use.

Live training dashboard: parallax.chutes.ai.

Model

Training

• Data: seven stationary sources totalling about 1.09T tokens: two general pretraining mixtures of web, document, math and code text, an additional web set, two math sets, a knowledge-focused set and a books set. At about 899.6B tokens the data switches to two further sources with no learning-rate event. Planned training totals about 972.9B tokens.
• Optimization: the dense trunk is synchronized with a decoupled DiLoCo scheme over the internet. Routed experts are trained through low-rank adapters on the GPUs that use them, and the adapter updates are folded into full-precision masters that publish new ternary versions. The tech report describes the details.
• Learning rate: linear warm-up from 0 to 6e-4 over the first 19.66B tokens, then 6e-4 until 400B tokens, 3e-4 until 700B and 1.5e-4 through the end of training. There is no final decay.
• Batch: 10 gradient-accumulation micro-batches per step at launch (about 8.5M tokens per fleet step with 26 machines), switched to 20 at about 100B tokens on 2026-10-04 at 22:08 UTC. This gives about 17M tokens per step with all 26 machines; tokens per step vary with fleet size.
• Fleet: 26 machines and 208 GPUs at launch. Three machines left on 2026-10-05 and their roles were reassigned automatically while training continued; the fleet has since comprised 23 machines and 184 GPUs.
• Document masking: In training, attention and recurrent state are reset at document boundaries (up to 64 documents per 4096-token sequence; further boundaries are merged into the 64th document).
• Output logits: the learnable output logit scale is capped at 3.0 in training and inference (see Files and format).

Exports

One folder per export under exports/, named by training tokens (rounded down to whole billions) and fleet step. A new export is added about every 30 minutes while the run continues. Published exports are never modified or removed.

• This table lists the exports only. Validation loss and benchmark results for this run are shown on the live training dashboard.

Latest export

exports/775B-tokens_step63160: 775.3B training tokens, step 63160.

hf download chutesai/parallax-8b-lambda --include "exports/775B-tokens_step63160/*" --local-dir parallax-8b-lambda
cd parallax-8b-lambda/exports/775B-tokens_step63160
tar -xf packed_experts.tar
sha256sum -c --quiet SHA256SUMS   # every file of the original export, byte for byte

Files and format

The files are in Parallax's native compact export format (inference only, no optimizer state). They are byte-identical to the export the training system produced:

The only change from the original export is packaging. The 4096 expert files are stored in one uncompressed tar to keep the repository's file count manageable. Extract it and check SHA256SUMS as shown above. Every upload was checked against the training system's own digests before and after it was published.

Logit-scale bound. The model's output logit scale is bounded: the forward pass uses exp(min(logit_scale_log, log 3.0)). The stored trunk tensor is the raw training parameter (it can sit slightly above the bound, e.g. from bf16 rounding), and each export records the bound in manifest.json (logit_scale: max, raw, effective) and in coverage.json (inference_policy.logit_scale_max). A loader must apply the recorded bound; the files themselves are left byte-identical.

Running it

This is a base model only. It is not chat- or instruction-tuned, so it does plain text completion: give it the start of a text and it continues it. It will not follow instructions or hold a conversation.

Standard transformers cannot load this format. Use the parallax-lambda branch of our llama.cpp fork, https://github.com/chutesai/llama.cpp. It adds a dedicated runtime, llama-parallax, for CPU (x86-64, ARM64) and Apple GPUs (Metal). Lambda's recurrent mixers (EDA) are implemented only on this branch. The fork's master branch runs the earlier kappa checkpoints and cannot run lambda.

A vLLM serving path for GPU servers is in preparation.

Ready-made GGUF

parallax-8b-lambda-725B-tokens.gguf at the repository root (2,133,006,592 bytes, 2.13 GB, sha256 eca13c0bf0bb4a5fbdf33b1616c7f33823ee84eb40fe7edb8852c566c7f6275f, also in the .sha256 file next to it) was converted from exports/725B-tokens_step58814 (725.2B training tokens, step 58814). It keeps the export's bf16 trunk and stores the routed experts' exact ternary codes and per-row scales; no weight is requantized. Each GGUF is a snapshot of one export and is not updated as new exports are added. To use a later export, convert it yourself (see tools/parallax/README.md on the branch).

The earlier snapshots remain at the repository root:

hf download chutesai/parallax-8b-lambda parallax-8b-lambda-725B-tokens.gguf --local-dir .
hf download NousResearch/Meta-Llama-3-8B tokenizer.json --local-dir .   # Llama 3 tokenizer

Build and run

Requires CMake, a C++17 compiler, and Python 3 with tokenizers for the text front end.

git clone -b parallax-lambda https://github.com/chutesai/llama.cpp && cd llama.cpp
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DGGML_METAL=ON   # -DGGML_METAL=OFF for CPU only
cmake --build build --target llama-parallax -j 8
build/bin/llama-parallax --backend metal --self-test               # or --backend cpu
pip install tokenizers

# Apple GPU
python tools/parallax/run.py --binary build/bin/llama-parallax \
  --model ../parallax-8b-lambda-725B-tokens.gguf --tokenizer ../tokenizer.json \
  --backend metal --experts lut9 --prompt 'The capital of France is' --predict 64

# CPU (x86-64 or ARM64)
python tools/parallax/run.py --binary build/bin/llama-parallax \
  --model ../parallax-8b-lambda-725B-tokens.gguf --tokenizer ../tokenizer.json \
  --backend cpu --experts lut8 --threads 8 --prompt 'The capital of France is' --predict 64

Generation is greedy by default. build/bin/llama-parallax --help lists sampling, context size (-c, default 4096), prompt scoring and benchmark options.

Speed and memory

Measured on an Apple M5 Max on AC power (batch 1, 128 generated tokens) with the bf16-trunk GGUF of the earlier 270B-tokens_step23035 export, which has the same architecture, using the same runtime code; speed does not depend on the weight values. Prompt times used --prefill-batch 128:

Peak resident memory was about 2.8 GB on Metal and 3.4 GB on CPU at the default 4,096-token context. Other Apple and ARM hardware will differ. ARM64 CPUs use NEON kernels. x86-64 CPUs use portable C++ that gives the same results but is not tuned for speed: on a shared, loaded 64-core AMD EPYC 9534 server it generated about 22 tokens/s with 16 threads.

Fidelity

The 725B GGUF was checked on CPU with lut against the training code's own model loaded from the same export, with FP32 activations in both, on the same 2,555 held-out positions from public text and code passages as before. Mean KL was 1.0e-5 nats/token, the top-1 next token agreed at all 2,555 positions, and three 48-token greedy continuations matched exactly.

The remaining differences are confined to two passages. Each begins at a token where the router's 12th- and 13th-ranked expert-selection scores were within a few FP32 rounding steps (in one case an exact tie in the runtime), so the two implementations picked different experts; the outputs diverge from that point. Excluding the positions from those tokens onward in those passages (2,068 positions remain), mean KL was 1.6e-9 nats/token with top-1 agreement at every position.

The training code itself runs in bf16; against that bf16 forward the top-1 agreement was 94.9% (mean KL 0.0027, measured on the earlier 270B-tokens_step23035 export), which reflects the bf16 rounding of the reference, not an error of the runtime.

On the 725B GGUF, Metal and CPU had per-layer relative error within 2.0e-5 at the final position of a 512-token text, and identical 48-token greedy continuations on three prompts. lut8 (CPU) quantizes expert lookup tables to 8 bits and is slightly approximate; lut and lut9 are exact.

Limitations of the runtime

• One sequence per process, greedy or simple sampled generation. No chat template, batching or server.
• Not integrated into llama-cli, llama-server or the standard llama.cpp model loader.
• Context is limited by -c. Inference uses a 2,048-token window for all sliding-window layers and dense global attention, the settings the run is evaluated with. Lambda was trained on 4,096-token sequences; longer inputs run but were not qualified.

Tech report

The Parallax tech report: https://parallax.chutes.ai/tech-report.pdf. It is AI-generated from the project's measurements and logs, and it is a living document that changes as the run progresses. The report currently published there describes an earlier run of the same system; lambda's architecture and training recipe are as described on this card.

Limitations

This is an early base model. It can produce incorrect, biased or nonsensical output. It has no alignment or safety tuning, and it is small and far from converged.

License

MIT.

_{Table updated 2026-10-09 03:30 UTC.}