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": 77440.47909631982,
-      "required_pages_per_second_upper_estimate": 0.0053575120305509
+      "required_token_positions_per_second": 77149.9655307806,
+      "required_pages_per_second_upper_estimate": 0.005337413628002554
     },
     {
       "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": 9680.059887039977,
-      "required_pages_per_second_upper_estimate": 0.004052838932066791
+      "required_token_positions_per_second": 9643.745691347574,
+      "required_pages_per_second_upper_estimate": 0.004037634936657021
     },
     {
       "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": 19360.119774079954,
-      "required_pages_per_second_upper_estimate": 0.004032202388248692
+      "required_token_positions_per_second": 19287.49138269515,
+      "required_pages_per_second_upper_estimate": 0.004017075809662719
     },
     {
       "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": 29040.179661119928,
-      "required_pages_per_second_upper_estimate": 0.0017409144288972698
+      "required_token_positions_per_second": 28931.23707404272,
+      "required_pages_per_second_upper_estimate": 0.0017343834871476902
     },
     {
       "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": 29040.179661119928,
-      "required_pages_per_second_upper_estimate": 0.0022577462344011167
+      "required_token_positions_per_second": 28931.23707404272,
+      "required_pages_per_second_upper_estimate": 0.0022492764274436626
     },
     {
       "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": 29040.179661119928,
-      "required_pages_per_second_upper_estimate": 0.3724962437772723
+      "required_token_positions_per_second": 28931.23707404272,
+      "required_pages_per_second_upper_estimate": 0.37109884524368236
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11726076431290267,
-    "producer_p95_seconds": 0.29149124400282744,
-    "gpu_update_seconds": 0.9520602256127282,
-    "producer_batches_per_second": 8.528001722140967,
-    "gpu_batches_per_second": 1.050353720381942,
-    "producer_headroom": 8.119171243607264,
+    "producer_batch_seconds": 0.1710767573668477,
+    "producer_p95_seconds": 0.28860257400083356,
+    "gpu_update_seconds": 0.9556452746642469,
+    "producer_batches_per_second": 5.84532940296299,
+    "gpu_batches_per_second": 1.0464133779674016,
+    "producer_headroom": 5.586061422797564,
     "sustainable": true,
-    "buffer_seconds": 60.931854439214604,
+    "buffer_seconds": 60.20565230384755,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334223.3253106609,
-    "upload_bytes_per_second": 5014097.910199312,
+    "generation_bytes_per_second": 326344.9628207539,
+    "upload_bytes_per_second": 5099719.624361744,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.71654529085035,
+    "estimated_full_shards_per_hour": 70.02603209940875,
     "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": 1.022619493000093,
-      "bytes": 1029637,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.0783612599989283,
       "bytes": 5165514,
@@ -562,6 +556,12 @@
       "bytes": 910531,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.064887002001342,
+      "bytes": 7934391,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -321,3 +321,4 @@
 {"step": 717100, "loss": 0.5374506115913391, "lr": 8e-05, "grad_norm": 0.21293672567364158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104963669682, "tokens_per_second": 152505.2003777048, "data_wait_s": 1.4142999134492129e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 717200, "loss": 0.569522500038147, "lr": 8e-05, "grad_norm": 0.28050026082668217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104978318972, "tokens_per_second": 152890.04641238428, "data_wait_s": 1.267300103791058e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 717300, "loss": 0.545729398727417, "lr": 8e-05, "grad_norm": 0.20329128435782476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104993011784, "tokens_per_second": 152404.97335077974, "data_wait_s": 1.9231003534514457e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 717400, "loss": 0.5065453052520752, "lr": 8e-05, "grad_norm": 0.20497184359867668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105007605613, "tokens_per_second": 151603.46260253413, "data_wait_s": 1.493200397817418e-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:ec7cc066fb5e083fcc7c36309b1e0570e49f444bde89eea78cff908683bc8c73
-size 688977
+oid sha256:29611ecdce643b4f30499038efdf21a63edfa08b44250621e7eb7ad6b53a3330
+size 691095