CyanMonkey commited on
Commit
ed7d71f
·
verified ·
1 Parent(s): feac368

round 8 step 4000 (14.04%)

Browse files
Files changed (5) hide show
  1. README.md +14 -14
  2. config.json +5 -5
  3. model.safetensors +1 -1
  4. progress.json +9 -9
  5. training_meta.json +32 -32
README.md CHANGED
@@ -13,11 +13,11 @@ tags:
13
 
14
  # Oleander-150M
15
 
16
- **Training checkpoint** (`round 7`) — step **3,500 / 28,500** · **12.28%** of the planned run.
17
 
18
  ```text
19
- steps ███░░░░░░░░░░░░░░░░░░░░░░░░░ 12.28% 3,500 / 28,500
20
- tokens ███░░░░░░░░░░░░░░░░░░░░░░░░░ 12.28% 1.84B / 14.94B compute
21
  phase tst (TST bag training ends at step 9,000)
22
  ```
23
 
@@ -27,17 +27,17 @@ Trained with the Oleander marimo trainer (molab): hybrid **GDN2** (gated delta-r
27
 
28
  | | |
29
  |---|---|
30
- | Step | **3,500** / 28,500 |
31
- | % complete (steps) | **12.28%** |
32
- | Compute tokens seen | **1.84B** / 14.94B (12.28%) |
33
- | Raw tokens seen (TST bags unpacked) | 11.01B / 38.54B |
34
- | Cluster tokens (all 1 worker(s)) | 1.84B / 14.94B compute |
35
  | Phase | `tst` |
36
- | Val loss / ppl | **6.0768** / **435.65** |
37
- | Train loss (latest) | 5.3462 |
38
- | Throughput | 76,478 compute tok/s · 0.142 steps/s |
39
- | Session elapsed | 0d 06h 56m · ETA 2d 00h 52m |
40
- | Checkpoint written | 2026-11-15T01:29:46Z |
41
 
42
  ## How to load
43
 
@@ -98,6 +98,6 @@ Oleander-150M (28L = 7 x [GDN2 · GDN2 · GDN2 · GQA] · d=576 · SwiGLU(1600)
98
 
99
  ## Notes
100
 
101
- - This is a **work-in-progress** checkpoint at 12.28% of the planned run, not a finished model: expect repetition and incoherence, especially with few-shot prompts.
102
  - The tokenizer comes from the base model the run started from; the vocabulary is a plain BPE model and is not extended during training.
103
  - `training_meta.json` is regenerated on every push and is the authoritative record of how far training got.
 
13
 
14
  # Oleander-150M
15
 
16
+ **Training checkpoint** (`round 8`) — step **4,000 / 28,500** · **14.04%** of the planned run.
17
 
18
  ```text
19
+ steps ████░░░░░░░░░░░░░░░░░░░░░░░░ 14.04% 4,000 / 28,500
20
+ tokens ████░░░░░░░░░░░░░░░░░░░░░░░░ 14.04% 2.10B / 14.94B compute
21
  phase tst (TST bag training ends at step 9,000)
22
  ```
23
 
 
27
 
28
  | | |
29
  |---|---|
30
+ | Step | **4,000** / 28,500 |
31
+ | % complete (steps) | **14.04%** |
32
+ | Compute tokens seen | **2.10B** / 14.94B (14.04%) |
33
+ | Raw tokens seen (TST bags unpacked) | 12.58B / 38.54B |
34
+ | Cluster tokens (all 1 worker(s)) | 2.10B / 14.94B compute |
35
  | Phase | `tst` |
36
+ | Val loss / ppl | **6.2148** / **500.08** |
37
+ | Train loss (latest) | 5.3614 |
38
+ | Throughput | 76,517 compute tok/s · 0.142 steps/s |
39
+ | Session elapsed | 0d 07h 53m · ETA 1d 23h 47m |
40
+ | Checkpoint written | 2026-11-15T02:27:25Z |
41
 
42
  ## How to load
43
 
 
98
 
99
  ## Notes
100
 
101
+ - This is a **work-in-progress** checkpoint at 14.04% of the planned run, not a finished model: expect repetition and incoherence, especially with few-shot prompts.
102
  - The tokenizer comes from the base model the run started from; the vocabulary is a plain BPE model and is not extended during training.
103
  - `training_meta.json` is regenerated on every push and is the authoritative record of how far training got.
config.json CHANGED
@@ -113,12 +113,12 @@
113
  },
114
  "model_name": "Oleander-150M",
115
  "training": {
116
- "step": 3500,
117
- "kind": "round 7",
118
  "max_steps": 28500,
119
- "percent_complete": 12.281,
120
- "tokens_seen": 11010048000,
121
- "compute_tokens_seen": 1835008000
122
  },
123
  "transformers_version": "5.14.1",
124
  "torch_dtype": "float32"
 
113
  },
114
  "model_name": "Oleander-150M",
115
  "training": {
116
+ "step": 4000,
117
+ "kind": "round 8",
118
  "max_steps": 28500,
119
+ "percent_complete": 14.035,
120
+ "tokens_seen": 12582912000,
121
+ "compute_tokens_seen": 2097152000
122
  },
123
  "transformers_version": "5.14.1",
124
  "torch_dtype": "float32"
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2179ada9d307145bba4cb7f58d2ddebdf47fde5fce0d41b73b7db6996f9cf2c7
3
  size 599970860
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a756b1b4e6896405c64b34f94b5ec74516153c85c58641b42a88a378b44099b8
3
  size 599970860
progress.json CHANGED
@@ -1,15 +1,15 @@
1
  {
2
- "step": 3500,
3
  "max_steps": 28500,
4
- "percent_complete": 12.281,
5
- "tokens_seen": 11010048000,
6
- "compute_tokens_seen": 1835008000,
7
  "compute_tokens_total": 14942208000,
8
- "percent_tokens": 12.281,
9
- "val_loss": 6.076828733086586,
10
- "val_ppl": 435.64545375862866,
11
  "params_M": 149.976,
12
- "pushed_at": "2026-11-15T01:29:46Z",
13
  "architecture": "Oleander-150M",
14
  "features": [
15
  "GDN2 (gated delta net 2) x21: 9h x 64, conv4",
@@ -25,5 +25,5 @@
25
  ],
26
  "optimizer": "muon_adamw",
27
  "phase": "tst",
28
- "kind": "round 7"
29
  }
 
1
  {
2
+ "step": 4000,
3
  "max_steps": 28500,
4
+ "percent_complete": 14.035,
5
+ "tokens_seen": 12582912000,
6
+ "compute_tokens_seen": 2097152000,
7
  "compute_tokens_total": 14942208000,
8
+ "percent_tokens": 14.035,
9
+ "val_loss": 6.214770317077637,
10
+ "val_ppl": 500.0811159068013,
11
  "params_M": 149.976,
12
+ "pushed_at": "2026-11-15T02:27:25Z",
13
  "architecture": "Oleander-150M",
14
  "features": [
15
  "GDN2 (gated delta net 2) x21: 9h x 64, conv4",
 
25
  ],
26
  "optimizer": "muon_adamw",
27
  "phase": "tst",
28
+ "kind": "round 8"
29
  }
training_meta.json CHANGED
@@ -1,10 +1,10 @@
1
  {
2
  "format": "oleander-training-meta/1",
3
- "generated_at": "2026-11-15T01:29:46Z",
4
- "kind": "round 7",
5
- "step": 3500,
6
- "tokens_seen": 11010048000,
7
- "pushed_at": "2026-11-15T01:29:46Z",
8
  "params_M": 149.976,
9
  "architecture": "Oleander-150M",
10
  "features": [
@@ -20,55 +20,55 @@
20
  "TST (token superposition training, bag 6)"
21
  ],
22
  "optimizer": "muon_adamw",
23
- "val_loss": 6.076828733086586,
24
- "val_ppl": 435.64545375862866,
25
  "run": {
26
  "model_name": "Oleander-150M",
27
  "architecture_detail": "Oleander-150M (28L = 7 x [GDN2 \u00b7 GDN2 \u00b7 GDN2 \u00b7 GQA] \u00b7 d=576 \u00b7 SwiGLU(1600) \u00b7 tied 32,770 vocab)",
28
  "session_id": "20261114_181817_2de909",
29
  "repo_id": "CyanMonkey/Test-Subject-3",
30
  "started_at": "2026-11-14T18:33:29Z",
31
- "session_elapsed_s": 24977.6,
32
- "last_update_at": "2026-11-15T01:29:46Z",
33
- "steps_per_second": 0.1421,
34
- "compute_tokens_per_second": 76478.2,
35
- "eta": "2d 00h 52m",
36
- "checkpoint_kind": "round 7"
37
  },
38
  "progress": {
39
- "step": 3500,
40
  "max_steps": 28500,
41
- "step_fraction": 0.122807,
42
- "percent_complete": 12.281,
43
- "tokens_fraction": 0.122807,
44
- "percent_tokens": 12.281,
45
  "phase": "tst",
46
- "phase_fraction": 0.388889,
47
  "tst_switch_step": 9000,
48
  "tst_switch_applied": false
49
  },
50
  "tokens": {
51
- "compute_tokens_seen": 1835008000,
52
  "compute_tokens_total": 14942208000,
53
- "raw_tokens_seen": 11010048000,
54
  "raw_tokens_total": 38535168000,
55
  "raw_over_compute_ratio": 6.0,
56
- "samples_seen": 10752000,
57
- "rows_consumed": 10752000,
58
  "compute_tokens_per_step": 524288,
59
  "cluster": {
60
  "n_workers": 1,
61
- "compute_tokens_seen_all_workers": 1835008000,
62
- "raw_tokens_seen_all_workers": 11010048000,
63
  "compute_tokens_total_planned_all_workers": 14942208000,
64
  "raw_tokens_total_planned_all_workers": 38535168000
65
  }
66
  },
67
  "metrics": {
68
- "train_loss": 5.346194267272949,
69
- "val_loss": 6.076828733086586,
70
- "val_ppl": 435.64545375862866,
71
- "grad_norm": 0.12433089315891266,
72
  "lr_mult": 1.0
73
  },
74
  "optimization": {
@@ -151,7 +151,7 @@
151
  "doc_masking": "auto",
152
  "shuffle_block_rows": 192,
153
  "seed": 42,
154
- "epochs_consumed_this_worker": 0.7288
155
  },
156
  "hardware": {
157
  "device": "cuda",
@@ -165,7 +165,7 @@
165
  "peak_gb": 39.36
166
  },
167
  "checkpoint": {
168
- "kind": "round 7",
169
  "weights_file": "model.safetensors",
170
  "dtype": "float32",
171
  "config_file": "config.json",
@@ -196,6 +196,6 @@
196
  "support_repo_id": "CyanMonkey/TestSubject-3-support"
197
  },
198
  "extra": {
199
- "diloco_round": 7
200
  }
201
  }
 
1
  {
2
  "format": "oleander-training-meta/1",
3
+ "generated_at": "2026-11-15T02:27:25Z",
4
+ "kind": "round 8",
5
+ "step": 4000,
6
+ "tokens_seen": 12582912000,
7
+ "pushed_at": "2026-11-15T02:27:25Z",
8
  "params_M": 149.976,
9
  "architecture": "Oleander-150M",
10
  "features": [
 
20
  "TST (token superposition training, bag 6)"
21
  ],
22
  "optimizer": "muon_adamw",
23
+ "val_loss": 6.214770317077637,
24
+ "val_ppl": 500.0811159068013,
25
  "run": {
26
  "model_name": "Oleander-150M",
27
  "architecture_detail": "Oleander-150M (28L = 7 x [GDN2 \u00b7 GDN2 \u00b7 GDN2 \u00b7 GQA] \u00b7 d=576 \u00b7 SwiGLU(1600) \u00b7 tied 32,770 vocab)",
28
  "session_id": "20261114_181817_2de909",
29
  "repo_id": "CyanMonkey/Test-Subject-3",
30
  "started_at": "2026-11-14T18:33:29Z",
31
+ "session_elapsed_s": 28436.3,
32
+ "last_update_at": "2026-11-15T02:27:24Z",
33
+ "steps_per_second": 0.1424,
34
+ "compute_tokens_per_second": 76517.1,
35
+ "eta": "1d 23h 47m",
36
+ "checkpoint_kind": "round 8"
37
  },
38
  "progress": {
39
+ "step": 4000,
40
  "max_steps": 28500,
41
+ "step_fraction": 0.140351,
42
+ "percent_complete": 14.035,
43
+ "tokens_fraction": 0.140351,
44
+ "percent_tokens": 14.035,
45
  "phase": "tst",
46
+ "phase_fraction": 0.444444,
47
  "tst_switch_step": 9000,
48
  "tst_switch_applied": false
49
  },
50
  "tokens": {
51
+ "compute_tokens_seen": 2097152000,
52
  "compute_tokens_total": 14942208000,
53
+ "raw_tokens_seen": 12582912000,
54
  "raw_tokens_total": 38535168000,
55
  "raw_over_compute_ratio": 6.0,
56
+ "samples_seen": 12288000,
57
+ "rows_consumed": 12288000,
58
  "compute_tokens_per_step": 524288,
59
  "cluster": {
60
  "n_workers": 1,
61
+ "compute_tokens_seen_all_workers": 2097152000,
62
+ "raw_tokens_seen_all_workers": 12582912000,
63
  "compute_tokens_total_planned_all_workers": 14942208000,
64
  "raw_tokens_total_planned_all_workers": 38535168000
65
  }
66
  },
67
  "metrics": {
68
+ "train_loss": 5.361351490020752,
69
+ "val_loss": 6.214770317077637,
70
+ "val_ppl": 500.0811159068013,
71
+ "grad_norm": 0.10999000817537308,
72
  "lr_mult": 1.0
73
  },
74
  "optimization": {
 
151
  "doc_masking": "auto",
152
  "shuffle_block_rows": 192,
153
  "seed": 42,
154
+ "epochs_consumed_this_worker": 0.8329
155
  },
156
  "hardware": {
157
  "device": "cuda",
 
165
  "peak_gb": 39.36
166
  },
167
  "checkpoint": {
168
+ "kind": "round 8",
169
  "weights_file": "model.safetensors",
170
  "dtype": "float32",
171
  "config_file": "config.json",
 
196
  "support_repo_id": "CyanMonkey/TestSubject-3-support"
197
  },
198
  "extra": {
199
+ "diloco_round": 8
200
  }
201
  }