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": 77014.23375120752,
-      "required_pages_per_second_upper_estimate": 0.00532802338855057
+      "required_token_positions_per_second": 76822.77368468372,
+      "required_pages_per_second_upper_estimate": 0.005314777736899893
     },
     {
       "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": 9626.77921890094,
-      "required_pages_per_second_upper_estimate": 0.004030531428943849
+      "required_token_positions_per_second": 9602.846710585465,
+      "required_pages_per_second_upper_estimate": 0.004020511387479764
     },
     {
       "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": 19253.55843780188,
-      "required_pages_per_second_upper_estimate": 0.0040100084721133855
+      "required_token_positions_per_second": 19205.69342117093,
+      "required_pages_per_second_upper_estimate": 0.0040000394514345325
     },
     {
       "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": 28880.33765670282,
-      "required_pages_per_second_upper_estimate": 0.0017313321447970733
+      "required_token_positions_per_second": 28808.540131756396,
+      "required_pages_per_second_upper_estimate": 0.001727027992805008
     },
     {
       "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": 28880.33765670282,
-      "required_pages_per_second_upper_estimate": 0.0022453192216283615
+      "required_token_positions_per_second": 28808.540131756396,
+      "required_pages_per_second_upper_estimate": 0.002239737279867714
     },
     {
       "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": 28880.33765670282,
-      "required_pages_per_second_upper_estimate": 0.3704459621695825
+      "required_token_positions_per_second": 28808.540131756396,
+      "required_pages_per_second_upper_estimate": 0.36952502060974585
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.15043227447660001,
-    "producer_p95_seconds": 0.32048423000378534,
-    "gpu_update_seconds": 0.9573295273984856,
-    "producer_batches_per_second": 6.647509674897268,
-    "gpu_batches_per_second": 1.0445723978842165,
-    "producer_headroom": 6.363857295446262,
+    "producer_batch_seconds": 0.18863566509349994,
+    "producer_p95_seconds": 0.31108263600617647,
+    "gpu_update_seconds": 0.9597154133306081,
+    "producer_batches_per_second": 5.301224450341009,
+    "gpu_batches_per_second": 1.0419755545340132,
+    "producer_headroom": 5.087666814517347,
     "sustainable": true,
-    "buffer_seconds": 60.311760226104596,
+    "buffer_seconds": 60.46207103982831,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272785.6652641606,
-    "upload_bytes_per_second": 3718281.0341218067,
+    "generation_bytes_per_second": 272586.77356074844,
+    "upload_bytes_per_second": 3744768.1468371483,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.53345364040006,
+    "estimated_full_shards_per_hour": 58.490776110809705,
     "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": 1.9732827289990382,
-      "bytes": 10394345,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.4147246300017287,
       "bytes": 2402371,
@@ -562,6 +556,12 @@
       "bytes": 1223490,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.546362834997126,
+      "bytes": 11822887,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 832900, "loss": 0.524799108505249, "lr": 3e-05, "grad_norm": 0.2792230709649194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121913337115, "tokens_per_second": 150326.6721979659, "data_wait_s": 7.94499646872282e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 833000, "loss": 0.4919930696487427, "lr": 3e-05, "grad_norm": 0.29146035383127855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121927826045, "tokens_per_second": 150087.6001304146, "data_wait_s": 1.237700053025037e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 833100, "loss": 0.4636203348636627, "lr": 3e-05, "grad_norm": 0.28554774813717965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121942377792, "tokens_per_second": 150481.27337229016, "data_wait_s": 2.219800080638379e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 833200, "loss": 0.5130776166915894, "lr": 3e-05, "grad_norm": 0.29794373546723096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121956979522, "tokens_per_second": 150814.42181676612, "data_wait_s": 1.4227000065147877e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1feac58c611df4fb2f49de99fcc1ddef18515965d40f06f6be2b7386c0664a81
-size 705921
+oid sha256:cd092585b4ff08fc4b9ff529c0479e50cfdf2012bfaf1412cc4df880aa8c0a67
+size 708039