Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75552.771757313,
-      "required_pages_per_second_upper_estimate": 0.0065487227691745295
+      "required_token_positions_per_second": 76010.02604814956,
+      "required_pages_per_second_upper_estimate": 0.006588356412203837
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9444.096469664126,
-      "required_pages_per_second_upper_estimate": 0.0037673912835743282
+      "required_token_positions_per_second": 9501.253256018696,
+      "required_pages_per_second_upper_estimate": 0.003790191980221276
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18888.19293932825,
-      "required_pages_per_second_upper_estimate": 0.004369451015671441
+      "required_token_positions_per_second": 19002.50651203739,
+      "required_pages_per_second_upper_estimate": 0.004395895448867533
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28332.28940899238,
-      "required_pages_per_second_upper_estimate": 0.0017252877769732323
+      "required_token_positions_per_second": 28503.75976805609,
+      "required_pages_per_second_upper_estimate": 0.0017357294222047658
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28332.28940899238,
-      "required_pages_per_second_upper_estimate": 0.0021810980675009838
+      "required_token_positions_per_second": 28503.75976805609,
+      "required_pages_per_second_upper_estimate": 0.0021942983304020416
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28332.28940899238,
-      "required_pages_per_second_upper_estimate": 0.3517922123867586
+      "required_token_positions_per_second": 28503.75976805609,
+      "required_pages_per_second_upper_estimate": 0.3539213003719544
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.22890363539852387,
-    "producer_p95_seconds": 0.35882261200094945,
-    "gpu_update_seconds": 0.9758477192183702,
-    "producer_batches_per_second": 4.368650581975198,
-    "gpu_batches_per_second": 1.0247500509618193,
-    "producer_headroom": 4.263137706482503,
+    "producer_batch_seconds": 0.08965579945314062,
+    "producer_p95_seconds": 0.3192756200005533,
+    "gpu_update_seconds": 0.9699773021166446,
+    "producer_batches_per_second": 11.153768145502497,
+    "gpu_batches_per_second": 1.0309519592034175,
+    "producer_headroom": 10.818901934209082,
     "sustainable": true,
-    "buffer_seconds": 60.50255859153895,
+    "buffer_seconds": 60.13859273123197,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 167,
+    "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 577571.3660680001,
-    "upload_bytes_per_second": 5146839.074837373,
+    "generation_bytes_per_second": 582172.1376332454,
+    "upload_bytes_per_second": 5247227.308742408,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.96668499245644,
+    "estimated_full_shards_per_hour": 62.46029423116695,
     "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": [
-    {
-      "seconds": 0.9246157949964982,
-      "bytes": 1204930,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0427682160006952,
       "bytes": 5985007,
@@ -562,6 +556,12 @@
       "bytes": 2380066,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0937937710004917,
+      "bytes": 10154198,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -310,3 +310,4 @@
 {"step": 146000, "loss": 0.6049474477767944, "lr": 0.0003, "grad_norm": 0.14668302956766735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21366663299, "tokens_per_second": 149947.5760015606, "data_wait_s": 1.5970999811543152e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 146100, "loss": 0.6441834196448326, "lr": 0.0003, "grad_norm": 0.14358625311955592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21381439536, "tokens_per_second": 150884.7213961859, "data_wait_s": 8.031001925701275e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 146200, "loss": 0.616436455398798, "lr": 0.0003, "grad_norm": 0.1636305480061172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21396067541, "tokens_per_second": 150013.2628389867, "data_wait_s": 1.0538999049458653e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 146300, "loss": 0.5969559364020824, "lr": 0.0003, "grad_norm": 0.16884647873112507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21410563663, "tokens_per_second": 147842.31492261676, "data_wait_s": 1.2511001841630787e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37aed4d02ea87458963acfdabfa70c0c6f4ac1cfc298e20e814c0625fc909e7a
-size 665679
+oid sha256:4dc7bfabfd0944fb2f22fb37020dc9ff9d3cc9045891bff84d92f564594880ca
+size 667797