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": 76854.1175658296,
-      "required_pages_per_second_upper_estimate": 0.005316946179325389
+      "required_token_positions_per_second": 75278.53433249312,
+      "required_pages_per_second_upper_estimate": 0.005207943675386391
     },
     {
       "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": 9606.7646957287,
-      "required_pages_per_second_upper_estimate": 0.004022151766042402
+      "required_token_positions_per_second": 9409.81679156164,
+      "required_pages_per_second_upper_estimate": 0.003939693791307568
     },
     {
       "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": 19213.5293914574,
-      "required_pages_per_second_upper_estimate": 0.0040016714773967715
+      "required_token_positions_per_second": 18819.63358312328,
+      "required_pages_per_second_upper_estimate": 0.003919633368251785
     },
     {
       "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": 28820.294087186103,
-      "required_pages_per_second_upper_estimate": 0.0017277326244857683
+      "required_token_positions_per_second": 28229.450374684922,
+      "required_pages_per_second_upper_estimate": 0.0016923124460874342
     },
     {
       "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": 28820.294087186103,
-      "required_pages_per_second_upper_estimate": 0.0022406510982021875
+      "required_token_positions_per_second": 28229.450374684922,
+      "required_pages_per_second_upper_estimate": 0.0021947155984020574
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28820.294087186103,
-      "required_pages_per_second_upper_estimate": 0.36967578772958404
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28229.450374684922,
+      "required_pages_per_second_upper_estimate": 0.3620970789841706
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0857777121484844,
-    "producer_p95_seconds": 0.24523496799884015,
-    "gpu_update_seconds": 0.9593240067696839,
-    "producer_batches_per_second": 11.65804000774657,
-    "gpu_batches_per_second": 1.0424006831302843,
-    "producer_headroom": 11.183837651312714,
+    "producer_batch_seconds": 0.17110320267192947,
+    "producer_p95_seconds": 0.2710215100014466,
+    "gpu_update_seconds": 0.9794027029585265,
+    "producer_batches_per_second": 5.844425962717857,
+    "gpu_batches_per_second": 1.0210304678343793,
+    "producer_headroom": 5.724046585126858,
     "sustainable": true,
-    "buffer_seconds": 60.437412426490084,
+    "buffer_seconds": 60.72296758342864,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263559.5974107922,
+    "generation_bytes_per_second": 263577.2742741723,
     "upload_bytes_per_second": 3921184.8852883764,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.55375425093483,
+    "estimated_full_shards_per_hour": 56.55754729431989,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 820900, "loss": 0.5293641090393066, "lr": 3e-05, "grad_norm": 0.2794738462263193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120156806130, "tokens_per_second": 150516.99113490837, "data_wait_s": 2.1830001060152426e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821000, "loss": 0.5010907053947449, "lr": 3e-05, "grad_norm": 0.2722542619171817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120171463232, "tokens_per_second": 151998.30985808626, "data_wait_s": 1.1166997865075245e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821100, "loss": 0.5874857306480408, "lr": 3e-05, "grad_norm": 0.29996884821400993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120186205649, "tokens_per_second": 152240.61913553393, "data_wait_s": 1.7594000382814556e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 821200, "loss": 0.5458189249038696, "lr": 3e-05, "grad_norm": 0.2782306516437237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120200872674, "tokens_per_second": 151976.70244240065, "data_wait_s": 1.0406998626422137e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9fbcdaec8160065f060821bc6349b86771a6fd0f185ee07df0a7bb203c85464
-size 450195
+oid sha256:272fb10564d9dab3d9198f4b225e6a8e49ea5e618fe8221899191b4fe827de2b
+size 452313