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": 76966.34088214427,
-      "required_pages_per_second_upper_estimate": 0.005324710048742008
+      "required_token_positions_per_second": 76458.56180960752,
+      "required_pages_per_second_upper_estimate": 0.005289580714294141
     },
     {
       "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": 9620.792610268034,
-      "required_pages_per_second_upper_estimate": 0.004028024960923855
+      "required_token_positions_per_second": 9557.32022620094,
+      "required_pages_per_second_upper_estimate": 0.004001450399169064
     },
     {
       "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": 19241.585220536068,
-      "required_pages_per_second_upper_estimate": 0.00400751476671189
+      "required_token_positions_per_second": 19114.64045240188,
+      "required_pages_per_second_upper_estimate": 0.003981075519268201
     },
     {
       "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": 28862.3778308041,
-      "required_pages_per_second_upper_estimate": 0.0017302554806575112
+      "required_token_positions_per_second": 28671.960678602816,
+      "required_pages_per_second_upper_estimate": 0.0017188402631332007
     },
     {
       "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": 28862.3778308041,
-      "required_pages_per_second_upper_estimate": 0.0022439229241616615
+      "required_token_positions_per_second": 28671.960678602816,
+      "required_pages_per_second_upper_estimate": 0.0022291188281345483
     },
     {
       "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": 28862.3778308041,
-      "required_pages_per_second_upper_estimate": 0.37021559280671235
+      "required_token_positions_per_second": 28671.960678602816,
+      "required_pages_per_second_upper_estimate": 0.367773126032283
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1264412003284292,
-    "producer_p95_seconds": 0.3480586570003652,
-    "gpu_update_seconds": 0.9579252327052546,
-    "producer_batches_per_second": 7.908814511429141,
-    "gpu_batches_per_second": 1.043922809273875,
-    "producer_headroom": 7.576052981283455,
+    "producer_batch_seconds": 0.13427870907037232,
+    "producer_p95_seconds": 0.35899128999881214,
+    "gpu_update_seconds": 0.9642870367296864,
+    "producer_batches_per_second": 7.447197004820202,
+    "gpu_batches_per_second": 1.0370356148221505,
+    "producer_headroom": 7.181235531720269,
     "sustainable": true,
-    "buffer_seconds": 60.34928966043104,
+    "buffer_seconds": 60.750083313970244,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 486201.8888262812,
-    "upload_bytes_per_second": 5578446.735043732,
+    "generation_bytes_per_second": 483131.5918176076,
+    "upload_bytes_per_second": 5530289.75515618,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.32760714141203,
+    "estimated_full_shards_per_hour": 103.66879287620706,
     "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": 1.3036935650015948,
-      "bytes": 11727766,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 0.7984904570039362,
       "bytes": 1227271,
@@ -562,6 +556,12 @@
       "bytes": 11190563,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.2993007859986392,
+      "bytes": 7936907,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -418,3 +418,4 @@
 {"step": 598900, "loss": 0.6015879511833191, "lr": 0.0003, "grad_norm": 0.17408554008761198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87661158218, "tokens_per_second": 149188.59794796898, "data_wait_s": 1.1969001207035035e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599000, "loss": 0.555098295211792, "lr": 0.0003, "grad_norm": 0.16050930060758728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87675752927, "tokens_per_second": 151080.69406588204, "data_wait_s": 1.4387995179276913e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599100, "loss": 0.5877408385276794, "lr": 0.0003, "grad_norm": 0.1686098600321335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87690455460, "tokens_per_second": 152727.06789891224, "data_wait_s": 1.0186995496042073e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 599200, "loss": 0.6008965969085693, "lr": 0.0003, "grad_norm": 0.19879451628919603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87705151674, "tokens_per_second": 153247.47532828877, "data_wait_s": 1.5926998457871377e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:835f2c53c7af227768b2c54c4db85e54eb3a6ea3bbb7f85dc9ea322e0a25bbe5
-size 895989
+oid sha256:56c16b2946695e5e39db7a7701206f718e1f6836a7f1694e5e3c8846ad19ca7e
+size 898107