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": 76657.15316727746,
-      "required_pages_per_second_upper_estimate": 0.005303319725213183
+      "required_token_positions_per_second": 74566.98581391899,
+      "required_pages_per_second_upper_estimate": 0.005158717097851404
     },
     {
       "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": 9582.144145909682,
-      "required_pages_per_second_upper_estimate": 0.004011843655968724
+      "required_token_positions_per_second": 9320.873226739874,
+      "required_pages_per_second_upper_estimate": 0.0039024549780695347
     },
     {
       "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": 19164.288291819365,
-      "required_pages_per_second_upper_estimate": 0.003991415854917291
+      "required_token_positions_per_second": 18641.746453479747,
+      "required_pages_per_second_upper_estimate": 0.003882584170346115
     },
     {
       "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": 28746.432437729047,
-      "required_pages_per_second_upper_estimate": 0.0017233047313810262
+      "required_token_positions_per_second": 27962.61968021962,
+      "required_pages_per_second_upper_estimate": 0.001676316353381119
     },
     {
       "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": 28746.432437729047,
-      "required_pages_per_second_upper_estimate": 0.002234908679839963
+      "required_token_positions_per_second": 27962.61968021962,
+      "required_pages_per_second_upper_estimate": 0.0021739706855715686
     },
     {
       "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": 28746.432437729047,
-      "required_pages_per_second_upper_estimate": 0.3687283697968093
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 27962.61968021962,
+      "required_pages_per_second_upper_estimate": 0.3586744613360478
     }
   ],
   "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": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07536875569529755,
-    "producer_p95_seconds": 0.16106941199996072,
-    "gpu_update_seconds": 0.961788912759575,
-    "producer_batches_per_second": 13.268097513017487,
-    "gpu_batches_per_second": 1.0397291824988806,
-    "producer_headroom": 12.76110908143311,
+    "producer_batch_seconds": 0.3920682803515412,
+    "producer_p95_seconds": 0.16754024200054118,
+    "gpu_update_seconds": 0.9887485620511379,
+    "producer_batches_per_second": 2.5505761371549043,
+    "gpu_batches_per_second": 1.0113794733875732,
+    "producer_headroom": 2.521878488013858,
     "sustainable": true,
-    "buffer_seconds": 60.59270150385323,
+    "buffer_seconds": 60.31366228511941,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326067.8161344017,
-    "upload_bytes_per_second": 5609063.499692655,
+    "generation_bytes_per_second": 303548.0888486434,
+    "upload_bytes_per_second": 5905642.770484794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.96656287216223,
+    "estimated_full_shards_per_hour": 65.13435362905958,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 970766,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.1719487760001357,
+      "bytes": 12751643,
+      "pages": 48,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -33,3 +33,4 @@
 {"step": 603300, "loss": 0.5545015931129456, "lr": 0.0003, "grad_norm": 0.1697075137537699, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88304881610, "tokens_per_second": 151125.20079138578, "data_wait_s": 1.1192999863851583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 603400, "loss": 0.5984642505645752, "lr": 0.0003, "grad_norm": 0.17757429015141551, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88319406939, "tokens_per_second": 150195.53006902183, "data_wait_s": 1.5753999832668342e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 603500, "loss": 0.6747608184814453, "lr": 0.0003, "grad_norm": 0.2245971242190504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88334247204, "tokens_per_second": 147668.0634139631, "data_wait_s": 1.7825999748311006e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 603600, "loss": 0.6014395356178284, "lr": 0.0003, "grad_norm": 0.1580584142154489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88349061194, "tokens_per_second": 154145.45711414694, "data_wait_s": 1.6590000086580403e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:262f3737a576710cc1a17f0aac1daa50345a18d3d949f6ce9748f07daaf1e535
-size 74295
+oid sha256:32a66864e791a6bad929bb5d047b3d3224d8d632103bf77f5143d6dd4a974edd
+size 76413