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": 75098.27650459661,
-      "required_pages_per_second_upper_estimate": 0.006509328272576759
+      "required_token_positions_per_second": 75288.57051603614,
+      "required_pages_per_second_upper_estimate": 0.006525822475192575
     },
     {
       "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": 9387.284563074576,
-      "required_pages_per_second_upper_estimate": 0.0037447281646220583
+      "required_token_positions_per_second": 9411.071314504517,
+      "required_pages_per_second_upper_estimate": 0.0037542170554110887
     },
     {
       "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": 18774.569126149152,
-      "required_pages_per_second_upper_estimate": 0.004343166146203269
+      "required_token_positions_per_second": 18822.142629009035,
+      "required_pages_per_second_upper_estimate": 0.004354171438824848
     },
     {
       "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": 28161.853689223728,
-      "required_pages_per_second_upper_estimate": 0.0017149091358464352
+      "required_token_positions_per_second": 28233.21394351355,
+      "required_pages_per_second_upper_estimate": 0.001719254600934366
     },
     {
       "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": 28161.853689223728,
-      "required_pages_per_second_upper_estimate": 0.0021679774539969244
+      "required_token_positions_per_second": 28233.21394351355,
+      "required_pages_per_second_upper_estimate": 0.0021734709639099816
     },
     {
       "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": 28161.853689223728,
-      "required_pages_per_second_upper_estimate": 0.34967597115889254
+      "required_token_positions_per_second": 28233.21394351355,
+      "required_pages_per_second_upper_estimate": 0.35056202668976433
     }
   ],
   "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": 0.5046190943828464,
-    "producer_p95_seconds": 0.2640495209998335,
-    "gpu_update_seconds": 0.9817535558953508,
-    "producer_batches_per_second": 1.9816927483154574,
-    "gpu_batches_per_second": 1.018585564569724,
-    "producer_headroom": 1.945533902350731,
+    "producer_batch_seconds": 0.07977011278140367,
+    "producer_p95_seconds": 0.2210376319999341,
+    "gpu_update_seconds": 0.9792721457541322,
+    "producer_batches_per_second": 12.536023394379907,
+    "gpu_batches_per_second": 1.0211665922856463,
+    "producer_headroom": 12.276178528638411,
     "sustainable": true,
-    "buffer_seconds": 60.86872046551175,
+    "buffer_seconds": 60.7148730367562,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 220887.34600948807,
-    "upload_bytes_per_second": 6617748.141877275,
+    "generation_bytes_per_second": 221714.02868585457,
+    "upload_bytes_per_second": 6583733.866930622,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.397282459387604,
+    "estimated_full_shards_per_hour": 47.574669317548064,
     "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": 1.2120231849994525,
-      "bytes": 11658602,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 0.9783495700003186,
       "bytes": 7342912,
@@ -562,6 +556,12 @@
       "bytes": 16651954,
       "pages": 29,
       "shards": 5
+    },
+    {
+      "seconds": 0.9651243050011544,
+      "bytes": 7745172,
+      "pages": 44,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -112,3 +112,4 @@
 {"step": 11200, "loss": 0.7446037046611309, "lr": 0.0003, "grad_norm": 0.14735876864574618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1638338285, "tokens_per_second": 148852.62966130654, "data_wait_s": 1.3658000170835294e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11300, "loss": 0.6859790533781052, "lr": 0.0003, "grad_norm": 0.16650099415743275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1653152694, "tokens_per_second": 150670.7353186621, "data_wait_s": 1.068199890141841e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11400, "loss": 0.7066514007747173, "lr": 0.0003, "grad_norm": 0.15130700510484968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1667767034, "tokens_per_second": 147853.70048813766, "data_wait_s": 1.3994998880662024e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 11500, "loss": 0.6995341554284096, "lr": 0.0003, "grad_norm": 0.15938632569844022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1682329231, "tokens_per_second": 148173.18191002306, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70acf5a09093979ba699aebb4e19b4613e2950b324e3e223294436d3ba6a4bfa
-size 239134
+oid sha256:7fd3529d7aab8d263e0b1842c70fbe75dfab371da14ffa853b80e586289eda13
+size 241217