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": 76251.96108573297,
-      "required_pages_per_second_upper_estimate": 0.0052752875968367745
+      "required_token_positions_per_second": 75845.4656511612,
+      "required_pages_per_second_upper_estimate": 0.005247165299473743
     },
     {
       "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": 9531.495135716621,
-      "required_pages_per_second_upper_estimate": 0.003990637973072494
+      "required_token_positions_per_second": 9480.68320639515,
+      "required_pages_per_second_upper_estimate": 0.003969364079339337
     },
     {
       "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": 19062.990271433242,
-      "required_pages_per_second_upper_estimate": 0.003970318148679304
+      "required_token_positions_per_second": 18961.3664127903,
+      "required_pages_per_second_upper_estimate": 0.003949152578925354
     },
     {
       "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": 28594.48540714986,
-      "required_pages_per_second_upper_estimate": 0.0017141957389074857
+      "required_token_positions_per_second": 28442.04961918545,
+      "required_pages_per_second_upper_estimate": 0.0017050574461749907
     },
     {
       "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": 28594.48540714986,
-      "required_pages_per_second_upper_estimate": 0.002223095466556786
+      "required_token_positions_per_second": 28442.04961918545,
+      "required_pages_per_second_upper_estimate": 0.002211244254536724
     },
     {
       "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": 28594.48540714986,
-      "required_pages_per_second_upper_estimate": 0.3667793564365498
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28442.04961918545,
+      "required_pages_per_second_upper_estimate": 0.3648240738213395
     }
   ],
   "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": 0.5094228423435823,
-    "producer_p95_seconds": 0.27352561600127956,
-    "gpu_update_seconds": 0.9668997223180243,
-    "producer_batches_per_second": 1.9630058114385573,
-    "gpu_batches_per_second": 1.0342334131636959,
-    "producer_headroom": 1.8980297739886092,
+    "producer_batch_seconds": 0.1404301015938927,
+    "producer_p95_seconds": 0.2638170690006518,
+    "gpu_update_seconds": 0.972081842559974,
+    "producer_batches_per_second": 7.120980392735755,
+    "gpu_batches_per_second": 1.028719965971696,
+    "producer_headroom": 6.92217574100402,
     "sustainable": true,
-    "buffer_seconds": 60.91468250603553,
+    "buffer_seconds": 60.26907423871839,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268608.4051413104,
+    "generation_bytes_per_second": 266300.58679623273,
     "upload_bytes_per_second": 4984750.1514002085,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.63711085967526,
+    "estimated_full_shards_per_hour": 57.14190676608311,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -176,3 +176,4 @@
 {"step": 737600, "loss": 0.5424453020095825, "lr": 5e-05, "grad_norm": 0.2345768996902286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107964223157, "tokens_per_second": 149091.45465566753, "data_wait_s": 1.7407000996172428e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737700, "loss": 0.5151412487030029, "lr": 5e-05, "grad_norm": 0.22466648249232482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107978785707, "tokens_per_second": 149306.4756886534, "data_wait_s": 2.1768999431515113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737800, "loss": 0.6282327175140381, "lr": 5e-05, "grad_norm": 0.27595751261107293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107993394586, "tokens_per_second": 149611.35810698985, "data_wait_s": 2.9964001441840082e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 737900, "loss": 0.5347310304641724, "lr": 5e-05, "grad_norm": 0.2168170256007732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108008061585, "tokens_per_second": 150672.47344936585, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd61cad9e61d180ff0cf2e8a887ac5fc4969f20b79e4016dd2b3b549e2010ef8
-size 379518
+oid sha256:5942470efc3dcdf644427e468ab55cacdab28505155b66ef7166ed68be25ee0c
+size 381636