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": 76518.28869561259,
-      "required_pages_per_second_upper_estimate": 0.005293712758853441
+      "required_token_positions_per_second": 77523.5095762051,
+      "required_pages_per_second_upper_estimate": 0.005363256271806621
     },
     {
       "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": 9564.786086951573,
-      "required_pages_per_second_upper_estimate": 0.004004576199160454
+      "required_token_positions_per_second": 9690.438697025638,
+      "required_pages_per_second_upper_estimate": 0.00405718432307359
     },
     {
       "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": 19129.572173903147,
-      "required_pages_per_second_upper_estimate": 0.003984185403080939
+      "required_token_positions_per_second": 19380.877394051276,
+      "required_pages_per_second_upper_estimate": 0.004036525653073469
     },
     {
       "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": 28694.35826085472,
-      "required_pages_per_second_upper_estimate": 0.0017201829639900744
+      "required_token_positions_per_second": 29071.31609107691,
+      "required_pages_per_second_upper_estimate": 0.0017427810103306154
     },
     {
       "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": 28694.35826085472,
-      "required_pages_per_second_upper_estimate": 0.002230860141638081
+      "required_token_positions_per_second": 29071.31609107691,
+      "required_pages_per_second_upper_estimate": 0.002260166954875591
     },
     {
       "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": 28694.35826085472,
-      "required_pages_per_second_upper_estimate": 0.3680604181687603
+      "required_token_positions_per_second": 29071.31609107691,
+      "required_pages_per_second_upper_estimate": 0.3728956284690667
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05377906467191451,
-    "producer_p95_seconds": 0.18589656800031662,
-    "gpu_update_seconds": 0.9635343557314479,
-    "producer_batches_per_second": 18.594596356418936,
-    "gpu_batches_per_second": 1.0378457125598495,
-    "producer_headroom": 17.916532420368448,
+    "producer_batch_seconds": 0.12370245287513626,
+    "producer_p95_seconds": 0.19725022000056924,
+    "gpu_update_seconds": 0.9510405347106462,
+    "producer_batches_per_second": 8.083914075732903,
+    "gpu_batches_per_second": 1.0514798933404554,
+    "producer_headroom": 7.688129965139939,
     "sustainable": true,
-    "buffer_seconds": 60.702664411081216,
+    "buffer_seconds": 60.86659422148136,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335653.8306744274,
-    "upload_bytes_per_second": 3194362.6806885493,
+    "generation_bytes_per_second": 348099.97583953745,
+    "upload_bytes_per_second": 3178474.034776925,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.0234984414541,
+    "estimated_full_shards_per_hour": 74.69415146245568,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 15924428,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0855836930004443,
+      "bytes": 1537634,
+      "pages": 45,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 936000, "loss": 0.5576494932174683, "lr": 3e-05, "grad_norm": 0.4360252231470891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137003943976, "tokens_per_second": 146671.57886310955, "data_wait_s": 1.644500116526615e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 936100, "loss": 0.5475285649299622, "lr": 3e-05, "grad_norm": 0.36074128690408375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137018500491, "tokens_per_second": 151856.23704198038, "data_wait_s": 1.1781001376220956e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 936200, "loss": 0.5170531868934631, "lr": 3e-05, "grad_norm": 0.36084369628317864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137033148233, "tokens_per_second": 153537.59229177586, "data_wait_s": 1.9698998585226946e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 936300, "loss": 0.5013501048088074, "lr": 3e-05, "grad_norm": 0.330866853622343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137047836782, "tokens_per_second": 153380.1884403988, "data_wait_s": 1.5082001482369378e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d92fe1fada5637e6e649e5b3a39b3db8350bd0a6d756fa4d9d7391dff9fa6298
-size 238947
+oid sha256:fed8a9dbb4ccfcaff8347d3628d42cd7b40ceb20e187b2367618dc2ab87d04eb
+size 241065