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": 76408.18680280261,
-      "required_pages_per_second_upper_estimate": 0.005286095654437258
+      "required_token_positions_per_second": 77661.04826134893,
+      "required_pages_per_second_upper_estimate": 0.005372771517178588
     },
     {
       "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": 9551.023350350326,
-      "required_pages_per_second_upper_estimate": 0.003998814028744133
+      "required_token_positions_per_second": 9707.631032668616,
+      "required_pages_per_second_upper_estimate": 0.004064382394990511
     },
     {
       "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": 19102.046700700652,
-      "required_pages_per_second_upper_estimate": 0.003978452572908402
+      "required_token_positions_per_second": 19415.262065337232,
+      "required_pages_per_second_upper_estimate": 0.004043687073317572
     },
     {
       "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": 28653.070051050974,
-      "required_pages_per_second_upper_estimate": 0.001717707798855786
+      "required_token_positions_per_second": 29122.89309800585,
+      "required_pages_per_second_upper_estimate": 0.0017458729731424755
     },
     {
       "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": 28653.070051050974,
-      "required_pages_per_second_upper_estimate": 0.002227650164933482
+      "required_token_positions_per_second": 29122.89309800585,
+      "required_pages_per_second_upper_estimate": 0.002264176840301038
     },
     {
       "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": 28653.070051050974,
-      "required_pages_per_second_upper_estimate": 0.3675308173452236
+      "required_token_positions_per_second": 29122.89309800585,
+      "required_pages_per_second_upper_estimate": 0.37355720293487576
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1592824902735117,
-    "producer_p95_seconds": 0.20932622699911008,
-    "gpu_update_seconds": 0.9649227796790972,
-    "producer_batches_per_second": 6.278153978399329,
-    "gpu_batches_per_second": 1.0363523600640543,
-    "producer_headroom": 6.057933788090463,
+    "producer_batch_seconds": 0.06993774363290584,
+    "producer_p95_seconds": 0.19259066300037375,
+    "gpu_update_seconds": 0.9493562300612627,
+    "producer_batches_per_second": 14.298430976682212,
+    "gpu_batches_per_second": 1.0533453811489384,
+    "producer_headroom": 13.574304527814204,
     "sustainable": true,
-    "buffer_seconds": 60.79013511978312,
+    "buffer_seconds": 60.75879872392081,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351721.19824491156,
-    "upload_bytes_per_second": 3482066.8785633515,
+    "generation_bytes_per_second": 352982.2695379813,
+    "upload_bytes_per_second": 3494888.9041931555,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.47118149290571,
+    "estimated_full_shards_per_hour": 75.74177803616122,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 8331124,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.3131415360003302,
+      "bytes": 5988822,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -102,3 +102,4 @@
 {"step": 655200, "loss": 0.5469732880592346, "lr": 0.0001, "grad_norm": 0.18391246918862672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95903806950, "tokens_per_second": 150940.5391757868, "data_wait_s": 8.176999472198077e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 655300, "loss": 0.5676447153091431, "lr": 0.0001, "grad_norm": 0.18752091429680115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95918428680, "tokens_per_second": 152529.30261704922, "data_wait_s": 1.6912999853957444e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 655400, "loss": 0.5680849552154541, "lr": 0.0001, "grad_norm": 0.18475815235608337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95933072724, "tokens_per_second": 153350.9804860552, "data_wait_s": 1.1686001016641967e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 655500, "loss": 0.5729182958602905, "lr": 0.0001, "grad_norm": 0.18111353609334457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95947683842, "tokens_per_second": 152825.6094391479, "data_wait_s": 1.31070009956602e-05, "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 sha256:7b0e9d2ca3345ef232e63c829be5e580976d2f8ba10904e6883189b5bf01e3cf
-size 222003
+oid sha256:903c2bfb51c272f9b49e66b616fd26d8be1c2ebaa78f91a08434a7bbf6e2a6ee
+size 224121