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": 76526.60631864822,
-      "required_pages_per_second_upper_estimate": 0.00529428819131459
+      "required_token_positions_per_second": 77066.49389402104,
+      "required_pages_per_second_upper_estimate": 0.005331638866490646
     },
     {
       "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": 9565.825789831028,
-      "required_pages_per_second_upper_estimate": 0.004005011501044616
+      "required_token_positions_per_second": 9633.31173675263,
+      "required_pages_per_second_upper_estimate": 0.004033266457753866
     },
     {
       "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": 19131.651579662055,
-      "required_pages_per_second_upper_estimate": 0.003984618488462902
+      "required_token_positions_per_second": 19266.62347350526,
+      "required_pages_per_second_upper_estimate": 0.004012729574502239
     },
     {
       "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": 28697.477369493085,
-      "required_pages_per_second_upper_estimate": 0.0017203699497903412
+      "required_token_positions_per_second": 28899.93521025789,
+      "required_pages_per_second_upper_estimate": 0.001732506988208968
     },
     {
       "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": 28697.477369493085,
-      "required_pages_per_second_upper_estimate": 0.002231102638615205
+      "required_token_positions_per_second": 28899.93521025789,
+      "required_pages_per_second_upper_estimate": 0.0022468428452167397
     },
     {
       "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": 28697.477369493085,
-      "required_pages_per_second_upper_estimate": 0.3681004267453353
+      "required_token_positions_per_second": 28899.93521025789,
+      "required_pages_per_second_upper_estimate": 0.3706973385443733
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.45754416279696386,
-    "producer_p95_seconds": 0.2071092870010034,
-    "gpu_update_seconds": 0.9634296298597753,
-    "producer_batches_per_second": 2.1855813740186476,
-    "gpu_batches_per_second": 1.0379585275424292,
-    "producer_headroom": 2.1056538541992045,
+    "producer_batch_seconds": 0.07800281725782554,
+    "producer_p95_seconds": 0.19555653399947914,
+    "gpu_update_seconds": 0.9566803454350472,
+    "producer_batches_per_second": 12.820049777108226,
+    "gpu_batches_per_second": 1.0452812214358322,
+    "producer_headroom": 12.264689649258397,
     "sustainable": true,
-    "buffer_seconds": 60.696066681165846,
+    "buffer_seconds": 60.270861762407975,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293430.11779692705,
+    "generation_bytes_per_second": 293180.42477170256,
     "upload_bytes_per_second": 3311001.481954556,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.96327257567271,
+    "estimated_full_shards_per_hour": 62.90969426501567,
     "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": [

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

@@ -117,3 +117,4 @@
 {"step": 696700, "loss": 0.5734730362892151, "lr": 8e-05, "grad_norm": 0.22531892352341243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101978008930, "tokens_per_second": 151518.06675428574, "data_wait_s": 2.0571000277413987e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696800, "loss": 0.6192939877510071, "lr": 8e-05, "grad_norm": 0.21230961375212332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101992517449, "tokens_per_second": 150895.99518598872, "data_wait_s": 1.5876999896136113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696900, "loss": 0.5699859261512756, "lr": 8e-05, "grad_norm": 0.2056741220829314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102007193347, "tokens_per_second": 153544.54882862, "data_wait_s": 1.6293000953737646e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 697000, "loss": 0.5464499592781067, "lr": 8e-05, "grad_norm": 0.20219522261892572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102021823622, "tokens_per_second": 152000.86391644052, "data_wait_s": 1.6635000065434724e-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:cf8b1b118fd14e5f5fcc096c9ce2b42c188455c9d4bbf36ef254dc4e67bd6648
-size 253773
+oid sha256:0bac4d218ee14889f8a76ffabe7a7589b88ee88d42fbbea7d9a977e5d981a7a4
+size 255891