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": 77551.03479212633,
-      "required_pages_per_second_upper_estimate": 0.005365160530111351
+      "required_token_positions_per_second": 76016.26946790084,
+      "required_pages_per_second_upper_estimate": 0.005258981903835249
     },
     {
       "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": 9693.879349015791,
-      "required_pages_per_second_upper_estimate": 0.004058624852212883
+      "required_token_positions_per_second": 9502.033683487605,
+      "required_pages_per_second_upper_estimate": 0.003978303078249287
     },
     {
       "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": 19387.758698031583,
-      "required_pages_per_second_upper_estimate": 0.004037958847220377
+      "required_token_positions_per_second": 19004.06736697521,
+      "required_pages_per_second_upper_estimate": 0.003958046061582182
     },
     {
       "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": 29081.638047047374,
-      "required_pages_per_second_upper_estimate": 0.0017433997958303303
+      "required_token_positions_per_second": 28506.101050462814,
+      "required_pages_per_second_upper_estimate": 0.001708897231679196
     },
     {
       "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": 29081.638047047374,
-      "required_pages_per_second_upper_estimate": 0.0022609694415508083
+      "required_token_positions_per_second": 28506.101050462814,
+      "required_pages_per_second_upper_estimate": 0.0022162239715860675
     },
     {
       "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": 29081.638047047374,
-      "required_pages_per_second_upper_estimate": 0.373028027437403
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28506.101050462814,
+      "required_pages_per_second_upper_estimate": 0.3656456568086968
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,37 @@
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.046661716015629295,
-    "producer_p95_seconds": 0.11371654799995667,
-    "gpu_update_seconds": 0.9507029815608021,
-    "producer_batches_per_second": 21.43084492788587,
-    "gpu_batches_per_second": 1.0518532279748036,
-    "producer_headroom": 20.37436817030829,
+    "producer_batch_seconds": 0.047245914593753824,
+    "producer_p95_seconds": 0.12817525200000546,
+    "gpu_update_seconds": 0.969897635283627,
+    "producer_batches_per_second": 21.165851240229895,
+    "gpu_batches_per_second": 1.0310366410034293,
+    "producer_headroom": 20.528709066664,
     "sustainable": true,
-    "buffer_seconds": 60.844990819891336,
+    "buffer_seconds": 60.13365338758488,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 34,
+    "upload_pages_per_commit": 42,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 298472.6638538232,
+    "upload_bytes_per_second": 22272.904219528333,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 64.04528557501814,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 0.9475190029999112,
+      "bytes": 21104,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790911926322536473-f9bff804/train.jsonl

@@ -1,2 +1,3 @@
 {"step": 483300, "loss": 0.5667192339897156, "lr": 0.0003, "grad_norm": 0.1487788756301438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70738479781, "tokens_per_second": 151634.87429747212, "data_wait_s": 1.1928000049010734e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 483400, "loss": 0.5242839455604553, "lr": 0.0003, "grad_norm": 0.15922517566295763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70753143675, "tokens_per_second": 154362.54684854476, "data_wait_s": 1.4137999983176996e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 483500, "loss": 0.5839016437530518, "lr": 0.0003, "grad_norm": 0.1528308329657215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70767836907, "tokens_per_second": 152693.96277359722, "data_wait_s": 2.2276999970927136e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790911926.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.444.0.1790911926322536473-f9bff804

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4bbbff914607f5cb19c826cf9760235969f98651ee45b4838032dcd2232ce84b
-size 4401
+oid sha256:ac144d461101796d65edc79a715e2d67e57456c9a45f0844336de4a04db0d4f2
+size 6519