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": 76959.91088843411,
-      "required_pages_per_second_upper_estimate": 0.005324265206857498
+      "required_token_positions_per_second": 77304.56969406469,
+      "required_pages_per_second_upper_estimate": 0.0053481095027496295
     },
     {
       "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": 9619.988861054264,
-      "required_pages_per_second_upper_estimate": 0.004027688447912242
+      "required_token_positions_per_second": 9663.071211758086,
+      "required_pages_per_second_upper_estimate": 0.004045726128490145
     },
     {
       "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": 19239.977722108528,
-      "required_pages_per_second_upper_estimate": 0.004007179967182018
+      "required_token_positions_per_second": 19326.142423516172,
+      "required_pages_per_second_upper_estimate": 0.004025125802169248
     },
     {
       "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": 28859.96658316279,
-      "required_pages_per_second_upper_estimate": 0.0017301109300431768
+      "required_token_positions_per_second": 28989.21363527426,
+      "required_pages_per_second_upper_estimate": 0.0017378590934684377
     },
     {
       "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": 28859.96658316279,
-      "required_pages_per_second_upper_estimate": 0.002243735460263507
+      "required_token_positions_per_second": 28989.21363527426,
+      "required_pages_per_second_upper_estimate": 0.0022537838500674722
     },
     {
       "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": 28859.96658316279,
-      "required_pages_per_second_upper_estimate": 0.3701846639109656
+      "required_token_positions_per_second": 28989.21363527426,
+      "required_pages_per_second_upper_estimate": 0.3718425063207791
     }
   ],
   "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": 0.10953216847647695,
-    "producer_p95_seconds": 0.2094678550001845,
-    "gpu_update_seconds": 0.9580052672732523,
-    "producer_batches_per_second": 9.129737993042284,
-    "gpu_batches_per_second": 1.0438355969025894,
-    "producer_headroom": 8.746337086159238,
+    "producer_batch_seconds": 0.09257880593752077,
+    "producer_p95_seconds": 0.18793627000013657,
+    "gpu_update_seconds": 0.9537340456299145,
+    "producer_batches_per_second": 10.801608314919035,
+    "gpu_batches_per_second": 1.0485103311369453,
+    "producer_headroom": 10.301861597497455,
     "sustainable": true,
-    "buffer_seconds": 60.3543318382149,
+    "buffer_seconds": 60.08524487468461,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 248297.2513333754,
-    "upload_bytes_per_second": 5128051.11877952,
+    "generation_bytes_per_second": 247888.0236123692,
+    "upload_bytes_per_second": 5140511.247586042,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.27881007195422,
+    "estimated_full_shards_per_hour": 53.190999329360075,
     "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": 1.2943038749999687,
-      "bytes": 2497791,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.873195456000758,
       "bytes": 6027495,
@@ -562,6 +556,12 @@
       "bytes": 10355941,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 1.0539684499999566,
+      "bytes": 2291279,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -121,3 +121,4 @@
 {"step": 657100, "loss": 0.5608146786689758, "lr": 0.0001, "grad_norm": 0.1846081286414038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96181683029, "tokens_per_second": 151656.85201946693, "data_wait_s": 9.3039998319e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 657200, "loss": 0.5723109245300293, "lr": 0.0001, "grad_norm": 0.18259844057913233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96196414328, "tokens_per_second": 154223.65756405285, "data_wait_s": 1.490099930379074e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 657300, "loss": 0.5231503844261169, "lr": 0.0001, "grad_norm": 0.1778715153824788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96211010820, "tokens_per_second": 153098.66837463455, "data_wait_s": 9.149000106845051e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 657400, "loss": 0.5541929006576538, "lr": 0.0001, "grad_norm": 0.22244126197697556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96225596595, "tokens_per_second": 152390.4617840014, "data_wait_s": 9.798999599297531e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:884d9dcee6014c055530d92063dccbfbeff0eec2e28de5817b5c728f45210dba
-size 262245
+oid sha256:761c4866cccc3f172ea1f72ecedd15901886fbd33727e5ab9e8cc2d039d4c846
+size 264363