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": 77464.54718314201,
-      "required_pages_per_second_upper_estimate": 0.0053591771166429155
+      "required_token_positions_per_second": 76586.96814759907,
+      "required_pages_per_second_upper_estimate": 0.005298464162700149
     },
     {
       "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": 9683.068397892752,
-      "required_pages_per_second_upper_estimate": 0.004054098532735998
+      "required_token_positions_per_second": 9573.371018449883,
+      "required_pages_per_second_upper_estimate": 0.00400817053070504
     },
     {
       "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": 19366.136795785504,
-      "required_pages_per_second_upper_estimate": 0.004033455575190426
+      "required_token_positions_per_second": 19146.742036899766,
+      "required_pages_per_second_upper_estimate": 0.003987761432743398
     },
     {
       "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": 29049.205193678255,
-      "required_pages_per_second_upper_estimate": 0.0017414554957929514
+      "required_token_positions_per_second": 28720.113055349648,
+      "required_pages_per_second_upper_estimate": 0.0017217269245947513
     },
     {
       "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": 29049.205193678255,
-      "required_pages_per_second_upper_estimate": 0.0022584479298584045
+      "required_token_positions_per_second": 28720.113055349648,
+      "required_pages_per_second_upper_estimate": 0.002232862463626693
     },
     {
       "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": 29049.205193678255,
-      "required_pages_per_second_upper_estimate": 0.3726120136180687
+      "required_token_positions_per_second": 28720.113055349648,
+      "required_pages_per_second_upper_estimate": 0.36839077301919737
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.437215778976622,
-    "producer_p95_seconds": 0.26078100800077664,
-    "gpu_update_seconds": 0.9517644223194379,
-    "producer_batches_per_second": 2.287200160846597,
-    "gpu_batches_per_second": 1.0506801647019044,
-    "producer_headroom": 2.1768757398170866,
+    "producer_batch_seconds": 0.23123948086725932,
+    "producer_p95_seconds": 0.2601726749999216,
+    "gpu_update_seconds": 0.9626703051870491,
+    "producer_batches_per_second": 4.324521038749606,
+    "gpu_batches_per_second": 1.038777237244996,
+    "producer_headroom": 4.163087988160898,
     "sustainable": true,
-    "buffer_seconds": 60.91292302844403,
+    "buffer_seconds": 60.648229226784096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338025.76141816063,
-    "upload_bytes_per_second": 5527221.822973858,
+    "generation_bytes_per_second": 335767.9586801492,
+    "upload_bytes_per_second": 5591680.691984401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.53245956333747,
+    "estimated_full_shards_per_hour": 72.047987654718,
     "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": 0.9930732779994287,
-      "bytes": 6062259,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.0064647640010662,
       "bytes": 5698061,
@@ -562,6 +556,12 @@
       "bytes": 6506928,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1265610020018357,
+      "bytes": 11819263,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -177,3 +177,4 @@
 {"step": 502700, "loss": 0.537597119808197, "lr": 0.0003, "grad_norm": 0.14868456586983175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73578574033, "tokens_per_second": 152649.4251514213, "data_wait_s": 1.2917000276502222e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502800, "loss": 0.5820568203926086, "lr": 0.0003, "grad_norm": 0.1621224558029401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73592991407, "tokens_per_second": 149879.22073531325, "data_wait_s": 1.3794000551570207e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502900, "loss": 0.6077935695648193, "lr": 0.0003, "grad_norm": 0.15652726329904404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73607510736, "tokens_per_second": 150449.8674342424, "data_wait_s": 1.1202002497157082e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 503000, "loss": 0.6127749085426331, "lr": 0.0003, "grad_norm": 0.15825898452385917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73622088908, "tokens_per_second": 151817.96304782038, "data_wait_s": 8.316997991641983e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dbf2565ad65665584107efc02ee440590b96f782a4a8bac9267b61a63495712d
-size 381636
+oid sha256:3efd1dc0193343e636f37ecc9c077fb7b82e394fa9157e166d5c166c367f981a
+size 383754