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": 77205.74263279322,
-      "required_pages_per_second_upper_estimate": 0.005341272417340495
+      "required_token_positions_per_second": 77148.60520968835,
+      "required_pages_per_second_upper_estimate": 0.005337319517833013
     },
     {
       "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": 9650.717829099152,
-      "required_pages_per_second_upper_estimate": 0.004040554025138814
+      "required_token_positions_per_second": 9643.575651211044,
+      "required_pages_per_second_upper_estimate": 0.0040375637444026976
     },
     {
       "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": 19301.435658198305,
-      "required_pages_per_second_upper_estimate": 0.004019980034514752
+      "required_token_positions_per_second": 19287.151302422088,
+      "required_pages_per_second_upper_estimate": 0.0040170049799103606
     },
     {
       "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": 28952.15348729746,
-      "required_pages_per_second_upper_estimate": 0.0017356373941847978
+      "required_token_positions_per_second": 28930.726953633133,
+      "required_pages_per_second_upper_estimate": 0.0017343529062080441
     },
     {
       "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": 28952.15348729746,
-      "required_pages_per_second_upper_estimate": 0.0022509025865726394
+      "required_token_positions_per_second": 28930.726953633133,
+      "required_pages_per_second_upper_estimate": 0.0022492367678255995
     },
     {
       "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": 28952.15348729746,
-      "required_pages_per_second_upper_estimate": 0.37136713853461933
+      "required_token_positions_per_second": 28930.726953633133,
+      "required_pages_per_second_upper_estimate": 0.37109230196679277
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06996793991403649,
-    "producer_p95_seconds": 0.18814130499958992,
-    "gpu_update_seconds": 0.9549548710471694,
-    "producer_batches_per_second": 14.292260158418452,
-    "gpu_batches_per_second": 1.0471699033310713,
-    "producer_headroom": 13.64846345655509,
+    "producer_batch_seconds": 0.10499692635938374,
+    "producer_p95_seconds": 0.19692283800031873,
+    "gpu_update_seconds": 0.9556621250586291,
+    "producer_batches_per_second": 9.524088320235181,
+    "gpu_batches_per_second": 1.0463949274317539,
+    "producer_headroom": 9.101810483362023,
     "sustainable": true,
-    "buffer_seconds": 60.16215687597167,
+    "buffer_seconds": 60.206713878693634,
     "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": 340027.3878116766,
-    "upload_bytes_per_second": 4533879.065807922,
+    "generation_bytes_per_second": 383855.77712919406,
+    "upload_bytes_per_second": 4495276.56340264,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.9619619919083,
+    "estimated_full_shards_per_hour": 82.36651406676165,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 5864908,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.0337117700000817,
+      "bytes": 1192866,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 850000, "loss": 0.5116078853607178, "lr": 3e-05, "grad_norm": 0.2902432474844378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124416666044, "tokens_per_second": 152880.71914155624, "data_wait_s": 7.378999725915492e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 850100, "loss": 0.563275933265686, "lr": 3e-05, "grad_norm": 0.30711087724013664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124431441815, "tokens_per_second": 146640.6605470716, "data_wait_s": 1.2968000191904139e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 850200, "loss": 0.5152894854545593, "lr": 3e-05, "grad_norm": 0.2985612132745051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124446006405, "tokens_per_second": 151491.3248043859, "data_wait_s": 1.0863000170502346e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 850300, "loss": 0.5646606683731079, "lr": 3e-05, "grad_norm": 0.31378122123391483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124460470927, "tokens_per_second": 150344.83567860647, "data_wait_s": 2.0165000023553148e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:814d1ad55667feda1f4465a1d6436c3cf58343c7b1e49d1b12c7d10a31d2693b
-size 164817
+oid sha256:3a5b1103b8ebd9a47af45f7b32b1c8fc950cc9e73902c6f4aaa315b9a46f90b4
+size 166935