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": 76216.2133610989,
-      "required_pages_per_second_upper_estimate": 0.005272814486300441
+      "required_token_positions_per_second": 77452.36473632511,
+      "required_pages_per_second_upper_estimate": 0.005358334306704951
     },
     {
       "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": 9527.026670137362,
-      "required_pages_per_second_upper_estimate": 0.003988767119846631
+      "required_token_positions_per_second": 9681.545592040638,
+      "required_pages_per_second_upper_estimate": 0.0040534609657255204
     },
     {
       "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": 19054.053340274724,
-      "required_pages_per_second_upper_estimate": 0.003968456821601708
+      "required_token_positions_per_second": 19363.091184081277,
+      "required_pages_per_second_upper_estimate": 0.004032821254590617
     },
     {
       "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": 28581.080010412086,
-      "required_pages_per_second_upper_estimate": 0.0017133921058419118
+      "required_token_positions_per_second": 29044.636776121915,
+      "required_pages_per_second_upper_estimate": 0.0017411816261360185
     },
     {
       "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": 28581.080010412086,
-      "required_pages_per_second_upper_estimate": 0.0022220532559245323
+      "required_token_positions_per_second": 29044.636776121915,
+      "required_pages_per_second_upper_estimate": 0.0022580927554877446
     },
     {
       "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": 28581.080010412086,
-      "required_pages_per_second_upper_estimate": 0.36660740640079126
+      "required_token_positions_per_second": 29044.636776121915,
+      "required_pages_per_second_upper_estimate": 0.3725534148628406
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.35031686014838215,
-    "producer_p95_seconds": 0.27970894900136045,
-    "gpu_update_seconds": 0.967353227727148,
-    "producer_batches_per_second": 2.854558583267829,
-    "gpu_batches_per_second": 1.0337485536173352,
-    "producer_headroom": 2.7613664592603695,
+    "producer_batch_seconds": 0.1614447013594713,
+    "producer_p95_seconds": 0.28157481100060977,
+    "gpu_update_seconds": 0.951914124907559,
+    "producer_batches_per_second": 6.194071354335804,
+    "gpu_batches_per_second": 1.0505149296919096,
+    "producer_headroom": 5.896224012877546,
     "sustainable": true,
-    "buffer_seconds": 60.94325334681033,
+    "buffer_seconds": 60.922503994083776,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297545.7404293591,
-    "upload_bytes_per_second": 4650296.626901075,
+    "generation_bytes_per_second": 307251.18174292205,
+    "upload_bytes_per_second": 4802205.579318685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.84638938580112,
+    "estimated_full_shards_per_hour": 65.92895116058108,
     "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": 2.3035276259979582,
-      "bytes": 7703540,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.1081734670005972,
       "bytes": 5790944,
@@ -562,6 +556,12 @@
       "bytes": 5896433,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.4275493559980532,
+      "bytes": 16198066,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -294,3 +294,4 @@
 {"step": 714400, "loss": 0.51793372631073, "lr": 8e-05, "grad_norm": 0.22320855063949332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104568324228, "tokens_per_second": 150712.13782569245, "data_wait_s": 1.3154000043869019e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 714500, "loss": 0.6017884612083435, "lr": 8e-05, "grad_norm": 0.24516137493931217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104582936399, "tokens_per_second": 151798.14309129078, "data_wait_s": 1.0315001418348402e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 714600, "loss": 0.5326101779937744, "lr": 8e-05, "grad_norm": 0.21582666366473535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104597547987, "tokens_per_second": 152773.97176997786, "data_wait_s": 1.1317002645228058e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 714700, "loss": 0.5153366327285767, "lr": 8e-05, "grad_norm": 0.2069151599215396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104612159098, "tokens_per_second": 151318.54636267663, "data_wait_s": 1.5069999790284783e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebc889543aef8997db8b6905801b06ce9680d623aceb1956c06ed385cb7e20df
-size 631008
+oid sha256:c1a90468cace8500fdda3afe2ca30b12042b48e3ed7db2923b3eb980bce5dee7
+size 633126