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": 77154.8149550827,
-      "required_pages_per_second_upper_estimate": 0.005337749122427988
+      "required_token_positions_per_second": 77256.40416056558,
+      "required_pages_per_second_upper_estimate": 0.005344777299382726
     },
     {
       "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": 9644.351869385337,
-      "required_pages_per_second_upper_estimate": 0.004037888730742995
+      "required_token_positions_per_second": 9657.050520070698,
+      "required_pages_per_second_upper_estimate": 0.004043205390606975
     },
     {
       "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": 19288.703738770673,
-      "required_pages_per_second_upper_estimate": 0.004017328311461271
+      "required_token_positions_per_second": 19314.101040141395,
+      "required_pages_per_second_upper_estimate": 0.004022617899564916
     },
     {
       "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": 28933.05560815601,
-      "required_pages_per_second_upper_estimate": 0.001734492505491041
+      "required_token_positions_per_second": 28971.151560212093,
+      "required_pages_per_second_upper_estimate": 0.0017367762996476524
     },
     {
       "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": 28933.05560815601,
-      "required_pages_per_second_upper_estimate": 0.0022494178104720845
+      "required_token_positions_per_second": 28971.151560212093,
+      "required_pages_per_second_upper_estimate": 0.002252379603178061
     },
     {
       "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": 28933.05560815601,
-      "required_pages_per_second_upper_estimate": 0.37112217144669785
+      "required_token_positions_per_second": 28971.151560212093,
+      "required_pages_per_second_upper_estimate": 0.37161082541542684
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.17930101653904273,
-    "producer_p95_seconds": 0.20639324900002975,
-    "gpu_update_seconds": 0.9555852093342757,
-    "producer_batches_per_second": 5.5772132211099334,
-    "gpu_batches_per_second": 1.0464791524940686,
-    "producer_headroom": 5.329502463396226,
+    "producer_batch_seconds": 0.07687787735935103,
+    "producer_p95_seconds": 0.20161003999965033,
+    "gpu_update_seconds": 0.9543286514703385,
+    "producer_batches_per_second": 13.00764321738086,
+    "gpu_batches_per_second": 1.0478570442785045,
+    "producer_headroom": 12.41356661045037,
     "sustainable": true,
-    "buffer_seconds": 60.20186818805937,
+    "buffer_seconds": 60.122705042631324,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 446824.0251519266,
-    "upload_bytes_per_second": 4779730.723018778,
+    "generation_bytes_per_second": 446772.5461731995,
+    "upload_bytes_per_second": 4802594.46749234,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.87803426664685,
+    "estimated_full_shards_per_hour": 95.8669880761813,
     "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": 1.294827431000158,
-      "bytes": 6143505,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.2396700680001231,
       "bytes": 7470065,
@@ -562,6 +556,12 @@
       "bytes": 9924170,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.1737768090006284,
+      "bytes": 7614180,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -122,3 +122,4 @@
 {"step": 854700, "loss": 0.5163973569869995, "lr": 3e-05, "grad_norm": 0.2944249649509857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125104272160, "tokens_per_second": 152227.42691248658, "data_wait_s": 1.465599962102715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 854800, "loss": 0.5010818243026733, "lr": 3e-05, "grad_norm": 0.312108886562663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125118782654, "tokens_per_second": 150846.80249759907, "data_wait_s": 6.986998414504342e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 854900, "loss": 0.5196844339370728, "lr": 3e-05, "grad_norm": 0.292664025456817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125133293220, "tokens_per_second": 150949.88166583338, "data_wait_s": 1.313800021307543e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 855000, "loss": 0.5274161696434021, "lr": 3e-05, "grad_norm": 0.32599872351762194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125147920557, "tokens_per_second": 140852.68936032, "data_wait_s": 1.17069994303165e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa25b9882eb5d49c4d6d7ce7b74731008c3661da61b48a8e0580387aab1224f5
-size 264363
+oid sha256:3c01eeb3aa4600678c4c4e9e20f9139a9663185e6395a7f89bf71ea8137cb2d1
+size 266481