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": 75724.28852144313,
-      "required_pages_per_second_upper_estimate": 0.0065635894075840645
+      "required_token_positions_per_second": 75565.06174324367,
+      "required_pages_per_second_upper_estimate": 0.006549788034006315
     },
     {
       "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": 9465.536065180391,
-      "required_pages_per_second_upper_estimate": 0.003775943858776285
+      "required_token_positions_per_second": 9445.632717905459,
+      "required_pages_per_second_upper_estimate": 0.003768004115966754
     },
     {
       "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": 18931.072130360782,
-      "required_pages_per_second_upper_estimate": 0.004379370361868829
+      "required_token_positions_per_second": 18891.265435810918,
+      "required_pages_per_second_upper_estimate": 0.004370161783658614
     },
     {
       "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": 28396.608195541172,
-      "required_pages_per_second_upper_estimate": 0.0017292044536194606
+      "required_token_positions_per_second": 28336.898153716378,
+      "required_pages_per_second_upper_estimate": 0.001725568425346709
     },
     {
       "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": 28396.608195541172,
-      "required_pages_per_second_upper_estimate": 0.0021860495057353063
+      "required_token_positions_per_second": 28336.898153716378,
+      "required_pages_per_second_upper_estimate": 0.002181452861427659
     },
     {
       "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": 28396.608195541172,
-      "required_pages_per_second_upper_estimate": 0.3525908364545634
+      "required_token_positions_per_second": 28336.898153716378,
+      "required_pages_per_second_upper_estimate": 0.35184943757175435
     }
   ],
   "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": 0.08452634467195708,
-    "producer_p95_seconds": 0.3419329819989798,
-    "gpu_update_seconds": 0.9736374080176688,
-    "producer_batches_per_second": 11.830631075801927,
-    "gpu_batches_per_second": 1.027076395961414,
-    "producer_headroom": 11.518744975857073,
+    "producer_batch_seconds": 0.34566703904678775,
+    "producer_p95_seconds": 0.30789110000114306,
+    "gpu_update_seconds": 0.9756890062568112,
+    "producer_batches_per_second": 2.8929573463457854,
+    "gpu_batches_per_second": 1.0249167445643945,
+    "producer_headroom": 2.8226266783994607,
     "sustainable": true,
-    "buffer_seconds": 60.36551929709547,
+    "buffer_seconds": 60.492718387922295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385160.3956169664,
-    "upload_bytes_per_second": 2804114.173088469,
+    "generation_bytes_per_second": 356381.26388679404,
+    "upload_bytes_per_second": 2764991.2014656165,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.64645482427353,
+    "estimated_full_shards_per_hour": 76.47112309887758,
     "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": 1.2717098209977848,
-      "bytes": 11010766,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.1728442359999462,
       "bytes": 8097014,
@@ -562,6 +556,12 @@
       "bytes": 11807082,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 3.1277496100010467,
+      "bytes": 10196563,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 220700, "loss": 0.5254691783338785, "lr": 0.0003, "grad_norm": 0.1393049401913934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32303755260, "tokens_per_second": 143998.6885894685, "data_wait_s": 1.5251000149874017e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 220800, "loss": 0.5944326967000961, "lr": 0.0003, "grad_norm": 0.1556542267930113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32318517758, "tokens_per_second": 149848.53562544571, "data_wait_s": 1.0022999049397185e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 220900, "loss": 0.5657839942723513, "lr": 0.0003, "grad_norm": 0.13758471348440585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32333370010, "tokens_per_second": 150897.88842711048, "data_wait_s": 1.410699769621715e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 221000, "loss": 0.6313675027340651, "lr": 0.0003, "grad_norm": 0.16740018193347608, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32347900388, "tokens_per_second": 146695.52538510278, "data_wait_s": 1.0517000191612169e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0149611393cdbf5ce924013e90e1efb0f9a9727fc72a3e0b0c495fee693ff5b1
-size 552642
+oid sha256:1859d7212e4810dd36acca0f48e9315811a38b7688dbfa19550fd4c7b48bb4a8
+size 554760