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": 77391.84543214839,
-      "required_pages_per_second_upper_estimate": 0.005354147440817873
+      "required_token_positions_per_second": 76303.7220054884,
+      "required_pages_per_second_upper_estimate": 0.0052788685373673355
     },
     {
       "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": 9673.980679018548,
-      "required_pages_per_second_upper_estimate": 0.004050293694616519
+      "required_token_positions_per_second": 9537.96525068605,
+      "required_pages_per_second_upper_estimate": 0.0039933468750988285
     },
     {
       "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": 19347.961358037097,
-      "required_pages_per_second_upper_estimate": 0.004029670110825959
+      "required_token_positions_per_second": 19075.9305013721,
+      "required_pages_per_second_upper_estimate": 0.003973013257318756
     },
     {
       "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": 29021.942037055644,
-      "required_pages_per_second_upper_estimate": 0.0017398211111818555
+      "required_token_positions_per_second": 28613.89575205815,
+      "required_pages_per_second_upper_estimate": 0.0017153593594468573
     },
     {
       "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": 29021.942037055644,
-      "required_pages_per_second_upper_estimate": 0.0022563283393489493
+      "required_token_positions_per_second": 28613.89575205815,
+      "required_pages_per_second_upper_estimate": 0.0022246045354963207
     },
     {
       "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": 29021.942037055644,
-      "required_pages_per_second_upper_estimate": 0.3722623111178108
+      "required_token_positions_per_second": 28613.89575205815,
+      "required_pages_per_second_upper_estimate": 0.3670283314998288
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.04838679430470094,
-    "producer_p95_seconds": 0.14686349600015092,
-    "gpu_update_seconds": 0.9526585079902173,
-    "producer_batches_per_second": 20.666795855555296,
-    "gpu_batches_per_second": 1.0496940840948945,
-    "producer_headroom": 19.688398904691716,
+    "producer_batch_seconds": 0.298357196992157,
+    "producer_p95_seconds": 0.15021178899996812,
+    "gpu_update_seconds": 0.9662438222174388,
+    "producer_batches_per_second": 3.3516872060783145,
+    "gpu_batches_per_second": 1.0349354655692329,
+    "producer_headroom": 3.238547056878399,
     "sustainable": true,
-    "buffer_seconds": 60.017486003383695,
+    "buffer_seconds": 60.873360799698645,
     "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": 422284.6825794669,
-    "upload_bytes_per_second": 5368731.293036958,
+    "generation_bytes_per_second": 392397.72092673654,
+    "upload_bytes_per_second": 5448735.614626381,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.61246259725576,
+    "estimated_full_shards_per_hour": 84.19941635943958,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 603387,
       "pages": 17,
       "shards": 1
+    },
+    {
+      "seconds": 1.5595179319998351,
+      "bytes": 10763416,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -28,3 +28,4 @@
 {"step": 559900, "loss": 0.5928296446800232, "lr": 0.0003, "grad_norm": 0.17015354657335416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81951488575, "tokens_per_second": 151433.72141466296, "data_wait_s": 1.1948000064876396e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 560000, "loss": 0.5581508874893188, "lr": 0.0003, "grad_norm": 0.15723244372163023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81966036542, "tokens_per_second": 151563.19488607012, "data_wait_s": 8.724000053916825e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 560100, "loss": 0.5691901445388794, "lr": 0.0003, "grad_norm": 0.18749308518000832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81980714131, "tokens_per_second": 137527.78140626746, "data_wait_s": 8.190999778889818e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 560200, "loss": 0.5885102152824402, "lr": 0.0003, "grad_norm": 0.1723210634054678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81995415301, "tokens_per_second": 153730.4404726955, "data_wait_s": 1.071900032911799e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6c470788312dd737da7847d33e2d6560c2969c7bf158e6a6d1df5746bd21234d
-size 64488
+oid sha256:dfd047d8800129e0e8e67318f35fcd1e052c89b045dd0d5a5104cb97f8e1d95c
+size 66606