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": 77438.13907525588,
-      "required_pages_per_second_upper_estimate": 0.0053573501424641
+      "required_token_positions_per_second": 76726.35327548358,
+      "required_pages_per_second_upper_estimate": 0.005308107149265254
     },
     {
       "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": 9679.767384406985,
-      "required_pages_per_second_upper_estimate": 0.0040527164673225076
+      "required_token_positions_per_second": 9590.794159435447,
+      "required_pages_per_second_upper_estimate": 0.004015465235999139
     },
     {
       "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": 19359.53476881397,
-      "required_pages_per_second_upper_estimate": 0.004032080547079414
+      "required_token_positions_per_second": 19181.588318870894,
+      "required_pages_per_second_upper_estimate": 0.003995018994319731
     },
     {
       "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": 29039.302153220957,
-      "required_pages_per_second_upper_estimate": 0.0017408618236386035
+      "required_token_positions_per_second": 28772.382478306343,
+      "required_pages_per_second_upper_estimate": 0.0017248603966902175
     },
     {
       "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": 29039.302153220957,
-      "required_pages_per_second_upper_estimate": 0.0022576780120217212
+      "required_token_positions_per_second": 28772.382478306343,
+      "required_pages_per_second_upper_estimate": 0.0022369261813526818
     },
     {
       "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": 29039.302153220957,
-      "required_pages_per_second_upper_estimate": 0.3724849880481389
+      "required_token_positions_per_second": 28772.382478306343,
+      "required_pages_per_second_upper_estimate": 0.36906122905435207
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07420339976555113,
-    "producer_p95_seconds": 0.18899669999882462,
-    "gpu_update_seconds": 0.9520889949117928,
-    "producer_batches_per_second": 13.476471471112424,
-    "gpu_batches_per_second": 1.050321981814994,
-    "producer_headroom": 12.830800177888877,
+    "producer_batch_seconds": 0.10367272029682795,
+    "producer_p95_seconds": 0.21280757100066694,
+    "gpu_update_seconds": 0.9609214676902721,
+    "producer_batches_per_second": 9.645738986465052,
+    "gpu_batches_per_second": 1.0406677690359645,
+    "producer_headroom": 9.268797663831275,
     "sustainable": true,
-    "buffer_seconds": 60.93369567435474,
+    "buffer_seconds": 60.538052464487144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 228754.13441086793,
-    "upload_bytes_per_second": 5803965.370535351,
+    "generation_bytes_per_second": 220385.74164052907,
+    "upload_bytes_per_second": 5818674.46758374,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 49.08531212086228,
+    "estimated_full_shards_per_hour": 47.2896498385611,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 5084502,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1290311129996553,
+      "bytes": 7603249,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -78,3 +78,4 @@
 {"step": 492800, "loss": 0.5880060791969299, "lr": 0.0003, "grad_norm": 0.1524960584163102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72129413368, "tokens_per_second": 151991.147364005, "data_wait_s": 9.154999133897945e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 492900, "loss": 0.5948826670646667, "lr": 0.0003, "grad_norm": 0.15656816627424516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72144115352, "tokens_per_second": 126557.60718155648, "data_wait_s": 1.3593000403488986e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 493000, "loss": 0.5357719659805298, "lr": 0.0003, "grad_norm": 0.1703783142994748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72158834728, "tokens_per_second": 153040.50293927226, "data_wait_s": 1.570799940964207e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 493100, "loss": 0.582404375076294, "lr": 0.0003, "grad_norm": 0.1659191974143131, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72173501495, "tokens_per_second": 153322.7852714244, "data_wait_s": 1.0695999662857503e-05, "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:fe6e9d0acd5fd3b5b82a109c7042e9a0bad485beb2d00b37cdc7deeddaca971e
-size 170388
+oid sha256:8c35dca49ff7ad479d6af4ab87a59a512a6fafcb4c4416d6cb9c9b8420f318fb
+size 172506