2 days ago
-
Auto-sync beta checkpoint from H100 pretraining runner
-
Upload verified beta checkpoint to continuous branch
-
sync(pretrained): [all] 177 datasets | 6/6 HEALTHY | nano step 12785 loss 0.0460 (+99.6%) PPL 2.42 | Rec #7,400→#9,900 (0% Dup · Ep 1)
### HELIOS Pretraining Telemetry & Health Sync — Trained Topics (2026-11-10 03:39 UTC)
- **Active Category**: `all` · **Synced Datasets**: `177 External Datasets (120,616 records / 90.38M tokens)`
- **Trained Topics & Slice**: `Coding 64% · Math 18% · Agents 14% · Geography 1% · Tagalog 1% · History 1% · Science 1% · General Web 1% · Law 1%` · `Rec #7,400→#9,900 (0% Dup · Ep 1)`
- **Checkpoint Health**: `6/6 HEALTHY (SHA-256 Verified · 0 NaN/Inf)`
| Tier | Mode | Category | Steps (Prev→Total) | Loss (Init→Prev→Curr / Gain) | Held-Out PPL | Tok/s (Train/Infer) | Status, SHA-256 & Trained Topics |
| :--- | :--- | :--- | ---: | :--- | ---: | ---: | :--- |
| **nano** | `full_parameter` | `all` | 12685 → **12785** | 10.4375 → 1.9111 → 0.0460 (+99.6%) | 3.28 → 2.42 (+26.2%) | 1653.2 / 8765.3 | **HEALTHY** (`91939b57ef50`) · Coding 64% · Math 18% · Agents 14% · Geography 1% · Tagalog 1% · History 1% · Science 1% · General Web 1% · Law 1% |
| **alpha** | `full_parameter` | `all` | 3366 → **34…
3 days ago
-
sync(pretrained): [all] 177 datasets | 6/6 HEALTHY | nano step 12685 loss 1.9111 (+81.7%) PPL 3.28 | Rec #4,900→#7,400 (0% Dup · Ep 1)
### HELIOS Pretraining Telemetry & Health Sync — Trained Topics (2026-11-09 17:11 UTC)
- **Active Category**: `all` · **Synced Datasets**: `177 External Datasets (120,616 records / 90.38M tokens)`
- **Trained Topics & Slice**: `Coding 64% · Math 18% · Agents 15% · Geography 1% · Tagalog 1% · History 1% · Science 1% · General Web 1% · Law 1%` · `Rec #4,900→#7,400 (0% Dup · Ep 1)`
- **Checkpoint Health**: `6/6 HEALTHY (SHA-256 Verified · 0 NaN/Inf)`
| Tier | Mode | Category | Steps (Prev→Total) | Loss (Init→Prev→Curr / Gain) | Held-Out PPL | Tok/s (Train/Infer) | Status, SHA-256 & Trained Topics |
| :--- | :--- | :--- | ---: | :--- | ---: | ---: | :--- |
| **nano** | `full_parameter` | `all` | 6298 → **12685** | 10.4375 → 2.3318 → 1.9111 (+81.7%) | 8.38 → 3.28 (+60.9%) | 1790.1 / 10160.0 | **HEALTHY** (`54439a9f2b48`) · Coding 64% · Math 18% · Agents 15% · Geography 1% · Tagalog 1% · History 1% · Science 1% · General Web 1% · Law 1% |
| **alpha** | `full_parameter` | `all` | 1412 → **33… -
sync(pretrained): [all] 177 datasets | 6/6 HEALTHY | nano step 6298 loss 2.3318 (+77.7%) PPL 8.38 | Rec #2,400→#4,900 (0% Dup · Ep 1)
### HELIOS Pretraining Telemetry & Health Sync — Trained Topics (2026-11-09 12:52 UTC)
- **Active Category**: `all` · **Synced Datasets**: `177 External Datasets (120,616 records / 90.38M tokens)`
- **Trained Topics & Slice**: `Coding 62% · Math 19% · Agents 16% · Geography 1% · Tagalog 1%` · `Rec #2,400→#4,900 (0% Dup · Ep 1)`
- **Checkpoint Health**: `6/6 HEALTHY (SHA-256 Verified · 0 NaN/Inf)`
| Tier | Mode | Category | Steps (Prev→Total) | Loss (Init→Prev→Curr / Gain) | Held-Out PPL | Tok/s (Train/Infer) | Status, SHA-256 & Trained Topics |
| :--- | :--- | :--- | ---: | :--- | ---: | ---: | :--- |
| **nano** | `full_parameter` | `all` | 3186 → **6298** | 10.4375 → 1.2157 → 2.3318 (+77.7%) | 19.39 → 8.38 (+56.8%) | 1736.2 / 10190.9 | **HEALTHY** (`3f036411ee72`) · Coding 62% · Math 19% · Agents 16% · Geography 1% · Tagalog 1% |
| **alpha** | `full_parameter` | `all` | 628 → **1412** | 10.5483 → 3.4652 → 1.4382 (+86.4%) | 76.45 → 14.20 (+81.4%) | 432.9 / 3274.0 | **HEALTHY** (`e0f92… -
sync(pretrained): [all] 177 datasets | 6/6 HEALTHY | nano step 3186 loss 1.2157 (+88.3%) PPL 19.39 [beta/SHA256SUMS]
### HELIOS Pretraining Telemetry & Health Sync (2026-11-09 11:37 UTC)
- **Active Category**: `all` · **Synced Datasets**: `177 External Datasets (120,616 records / 90.38M tokens)`
- **Checkpoint Health**: `6/6 HEALTHY (SHA-256 Verified · 0 NaN/Inf)`
| Tier | Mode | Category | Steps (Prev→Total) | Loss (Init→Prev→Curr / Gain) | Held-Out PPL | Tok/s (Train/Infer) | Status & SHA-256 |
| :--- | :--- | :--- | ---: | :--- | ---: | ---: | :--- |
| **nano** | `full_parameter` | `all` | 516 → **3186** | 10.4375 → 2.8134 → 1.2157 (+88.3%) | 19.39 | 3416.6 / 9028.3 | **HEALTHY** (`b4a0af1fac86`) |
| **alpha** | `full_parameter` | `all` | 156 → **628** | 10.5483 → 4.0479 → 3.4652 (+67.2%) | 76.45 | 602.8 / 2736.1 | **HEALTHY** (`932d5cb4e15d`) |
| **beta** | `full_parameter` | `all` | 46 → **178** | 10.6157 → 5.9297 → 3.8918 (+63.3%) | 213.03 | 169.5 / 986.7 | **HEALTHY** (`5e9959304ea7`) |
| **core** | `block_slice` | `all` | 323 → **1793** | 2016.0848 → 19.7864 → 1.1983 (+99.9%) | 745.77 | 742.… -
sync(pretrained): [all] 177 datasets | 6/6 HEALTHY | nano step 3186 loss 1.2157 (+88.3%) PPL 19.39 [beta/manifest.json]
### HELIOS Pretraining Telemetry & Health Sync (2026-11-09 11:37 UTC)
- **Active Category**: `all` · **Synced Datasets**: `177 External Datasets (120,616 records / 90.38M tokens)`
- **Checkpoint Health**: `6/6 HEALTHY (SHA-256 Verified · 0 NaN/Inf)`
| Tier | Mode | Category | Steps (Prev→Total) | Loss (Init→Prev→Curr / Gain) | Held-Out PPL | Tok/s (Train/Infer) | Status & SHA-256 |
| :--- | :--- | :--- | ---: | :--- | ---: | ---: | :--- |
| **nano** | `full_parameter` | `all` | 516 → **3186** | 10.4375 → 2.8134 → 1.2157 (+88.3%) | 19.39 | 3416.6 / 9028.3 | **HEALTHY** (`b4a0af1fac86`) |
| **alpha** | `full_parameter` | `all` | 156 → **628** | 10.5483 → 4.0479 → 3.4652 (+67.2%) | 76.45 | 602.8 / 2736.1 | **HEALTHY** (`932d5cb4e15d`) |
| **beta** | `full_parameter` | `all` | 46 → **178** | 10.6157 → 5.9297 → 3.8918 (+63.3%) | 213.03 | 169.5 / 986.7 | **HEALTHY** (`5e9959304ea7`) |
| **core** | `block_slice` | `all` | 323 → **1793** | 2016.0848 → 19.7864 → 1.1983 (+99.9%) | 745.77 | 742.… -
chore: sync beta continuous checkpoint config.json
-
chore: sync beta continuous checkpoint manifest.json
