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": 76948.50974176789,
-      "required_pages_per_second_upper_estimate": 0.005323476449076823
+      "required_token_positions_per_second": 77144.54314927735,
+      "required_pages_per_second_upper_estimate": 0.005337038495068501
     },
     {
       "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": 9618.563717720986,
-      "required_pages_per_second_upper_estimate": 0.004027091770158975
+      "required_token_positions_per_second": 9643.067893659669,
+      "required_pages_per_second_upper_estimate": 0.00403735115691912
     },
     {
       "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": 19237.127435441973,
-      "required_pages_per_second_upper_estimate": 0.004006586327636489
+      "required_token_positions_per_second": 19286.135787319337,
+      "required_pages_per_second_upper_estimate": 0.0040167934748953915
     },
     {
       "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": 28855.69115316296,
-      "required_pages_per_second_upper_estimate": 0.0017298546245428906
+      "required_token_positions_per_second": 28929.203680979008,
+      "required_pages_per_second_upper_estimate": 0.001734261588338332
     },
     {
       "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": 28855.69115316296,
-      "required_pages_per_second_upper_estimate": 0.0022434030643866494
+      "required_token_positions_per_second": 28929.203680979008,
+      "required_pages_per_second_upper_estimate": 0.00224911834007691
     },
     {
       "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": 28855.69115316296,
-      "required_pages_per_second_upper_estimate": 0.37012982328552685
+      "required_token_positions_per_second": 28929.203680979008,
+      "required_pages_per_second_upper_estimate": 0.37107276306074843
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3792851949065721,
-    "producer_p95_seconds": 0.3432538499982911,
-    "gpu_update_seconds": 0.9581472110041426,
-    "producer_batches_per_second": 2.636538450298136,
-    "gpu_batches_per_second": 1.0436809589541,
-    "producer_headroom": 2.526191962858343,
+    "producer_batch_seconds": 0.11342319024225844,
+    "producer_p95_seconds": 0.33863598400057526,
+    "gpu_update_seconds": 0.9557124456273437,
+    "producer_batches_per_second": 8.816539173903669,
+    "gpu_batches_per_second": 1.0463398322113355,
+    "producer_headroom": 8.426076215860757,
     "sustainable": true,
-    "buffer_seconds": 60.36327429326098,
+    "buffer_seconds": 60.209884074522655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 552775.9186551522,
-    "upload_bytes_per_second": 3744767.1335329753,
+    "generation_bytes_per_second": 556548.065335322,
+    "upload_bytes_per_second": 3804439.899609041,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.61284417859005,
+    "estimated_full_shards_per_hour": 119.42225904507394,
     "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": 1.127826156996889,
-      "bytes": 2574706,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.1363400170012028,
       "bytes": 10297243,
@@ -562,6 +556,12 @@
       "bytes": 8343147,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.4054391019963077,
+      "bytes": 10539085,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -411,3 +411,4 @@
 {"step": 526100, "loss": 0.6048679351806641, "lr": 0.0003, "grad_norm": 0.15982237794437185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77003576077, "tokens_per_second": 150215.59958819396, "data_wait_s": 1.3611999747809023e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526200, "loss": 0.543147623538971, "lr": 0.0003, "grad_norm": 0.16020451243368156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77018180283, "tokens_per_second": 151837.3865375966, "data_wait_s": 1.4539000403601676e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526300, "loss": 0.638948380947113, "lr": 0.0003, "grad_norm": 0.16214304313669964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77032790042, "tokens_per_second": 152668.93492124518, "data_wait_s": 1.5463003364857286e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 526400, "loss": 0.539824366569519, "lr": 0.0003, "grad_norm": 0.15327712469377483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77047436084, "tokens_per_second": 153171.25112464483, "data_wait_s": 1.7212005332112312e-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 sha256:40dc817418e7eae4d24c0595fd1d79c113a2b4b4190deb73ee3a42be30086f0c
-size 881163
+oid sha256:80e2e371ad4e46d513b4725545339e8da9b05a69395bf8ac4b6f00be434a3c03
+size 883281