Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77459.78546178447,
-      "required_pages_per_second_upper_estimate": 0.005358847689710684
+      "required_token_positions_per_second": 76961.68739717902,
+      "required_pages_per_second_upper_estimate": 0.005324388109854539
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9682.473182723059,
-      "required_pages_per_second_upper_estimate": 0.004053849328573954
+      "required_token_positions_per_second": 9620.210924647377,
+      "required_pages_per_second_upper_estimate": 0.004027781421301463
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19364.946365446118,
-      "required_pages_per_second_upper_estimate": 0.00403320763994449
+      "required_token_positions_per_second": 19240.421849294755,
+      "required_pages_per_second_upper_estimate": 0.004007272467162488
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29047.419548169175,
-      "required_pages_per_second_upper_estimate": 0.001741348449071205
+      "required_token_positions_per_second": 28860.632773942132,
+      "required_pages_per_second_upper_estimate": 0.0017301508671631839
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29047.419548169175,
-      "required_pages_per_second_upper_estimate": 0.002258309103774298
+      "required_token_positions_per_second": 28860.632773942132,
+      "required_pages_per_second_upper_estimate": 0.002243787253666337
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29047.419548169175,
-      "required_pages_per_second_upper_estimate": 0.372589109274755
+      "required_token_positions_per_second": 28860.632773942132,
+      "required_pages_per_second_upper_estimate": 0.3701932090909831
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0839357115234236,
-    "producer_p95_seconds": 0.17797104399960517,
-    "gpu_update_seconds": 0.951822930575691,
-    "producer_batches_per_second": 11.913880061896348,
-    "gpu_batches_per_second": 1.050615579722554,
-    "producer_headroom": 11.339904235041477,
+    "producer_batch_seconds": 0.3579961058593639,
+    "producer_p95_seconds": 0.1912494259995583,
+    "gpu_update_seconds": 0.9579831536113442,
+    "producer_batches_per_second": 2.793326473760143,
+    "gpu_batches_per_second": 1.0438596923445504,
+    "producer_headroom": 2.6759597043987977,
     "sustainable": true,
-    "buffer_seconds": 60.916667556844224,
+    "buffer_seconds": 60.352938677514686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339725.0392755915,
-    "upload_bytes_per_second": 3999333.7876276877,
+    "generation_bytes_per_second": 318835.09161550604,
+    "upload_bytes_per_second": 4028460.163920026,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.89708503437814,
+    "estimated_full_shards_per_hour": 68.4145885596169,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -406,6 +406,12 @@
       "bytes": 6589194,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.0017207049995704,
+      "bytes": 5856566,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -62,3 +62,4 @@
 {"step": 531200, "loss": 0.5846951603889465, "lr": 0.0003, "grad_norm": 0.16805864125753955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77750483592, "tokens_per_second": 153595.62734738825, "data_wait_s": 1.3156999557395466e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 531300, "loss": 0.7076463103294373, "lr": 0.0003, "grad_norm": 0.19203421174432692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77765064236, "tokens_per_second": 152195.67498223743, "data_wait_s": 1.3093999768898357e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 531400, "loss": 0.5621283650398254, "lr": 0.0003, "grad_norm": 0.14917915873942325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77779609710, "tokens_per_second": 151766.83163119946, "data_wait_s": 9.91100023384206e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 531500, "loss": 0.5958731174468994, "lr": 0.0003, "grad_norm": 0.19584721023642038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77794232968, "tokens_per_second": 153315.87980441257, "data_wait_s": 1.239099947270006e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93a1865cbe066040ef1e0fc7801f3f9e6143d3b4c79d4bfe098f2c1a353c5a8c
-size 136500
+oid sha256:c17bea5d1d93f8aa0e604c37514e3f0dfa9c1f1a9309606093f9cc8d0a98dfb9
+size 138618