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": 76638.42162860332,
-      "required_pages_per_second_upper_estimate": 0.005302023833904547
+      "required_token_positions_per_second": 75635.97014906333,
+      "required_pages_per_second_upper_estimate": 0.005232671914542077
     },
     {
       "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": 9579.802703575415,
-      "required_pages_per_second_upper_estimate": 0.0040108633429582425
+      "required_token_positions_per_second": 9454.496268632916,
+      "required_pages_per_second_upper_estimate": 0.003958400155343734
     },
     {
       "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": 19159.60540715083,
-      "required_pages_per_second_upper_estimate": 0.003990440533536809
+      "required_token_positions_per_second": 18908.992537265833,
+      "required_pages_per_second_upper_estimate": 0.0039382444818455865
     },
     {
       "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": 28739.408110726243,
-      "required_pages_per_second_upper_estimate": 0.0017228836337027343
+      "required_token_positions_per_second": 28363.48880589875,
+      "required_pages_per_second_upper_estimate": 0.0017003478453738693
     },
     {
       "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": 28739.408110726243,
-      "required_pages_per_second_upper_estimate": 0.0022343625693122438
+      "required_token_positions_per_second": 28363.48880589875,
+      "required_pages_per_second_upper_estimate": 0.0022051364968561815
     },
     {
       "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": 28739.408110726243,
-      "required_pages_per_second_upper_estimate": 0.368638269272152
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28363.48880589875,
+      "required_pages_per_second_upper_estimate": 0.363816380060527
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.26477699482033756,
-    "producer_p95_seconds": 0.22772088300007454,
-    "gpu_update_seconds": 0.9620239878802896,
-    "producer_batches_per_second": 3.7767631613106816,
-    "gpu_batches_per_second": 1.039475119745596,
-    "producer_headroom": 3.6333367577234714,
+    "producer_batch_seconds": 0.4434987825702308,
+    "producer_p95_seconds": 0.2324553089983965,
+    "gpu_update_seconds": 0.9747743018922993,
+    "producer_batches_per_second": 2.2547976213252485,
+    "gpu_batches_per_second": 1.0258785013707592,
+    "producer_headroom": 2.1979187772357363,
     "sustainable": true,
-    "buffer_seconds": 60.60751123645825,
+    "buffer_seconds": 60.436006717322556,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379620.99676541204,
-    "upload_bytes_per_second": 5329666.288141534,
+    "generation_bytes_per_second": 325911.47470271605,
+    "upload_bytes_per_second": 5338589.877425503,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.45782878133555,
+    "estimated_full_shards_per_hour": 69.93301564036476,
     "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": 1.1522871959996337,
-      "bytes": 11018272,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 3.0842820179996124,
       "bytes": 6556297,
@@ -562,6 +556,12 @@
       "bytes": 7651274,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1375053659994592,
+      "bytes": 11652818,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -119,3 +119,4 @@
 {"step": 409000, "loss": 0.5870890617370605, "lr": 0.0003, "grad_norm": 0.1879548400610122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59866805781, "tokens_per_second": 152714.8024420476, "data_wait_s": 1.4307001038105227e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 409100, "loss": 0.6051620244979858, "lr": 0.0003, "grad_norm": 0.18104377254494006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59881548404, "tokens_per_second": 153038.28477918432, "data_wait_s": 9.593000868335366e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 409200, "loss": 0.6069282293319702, "lr": 0.0003, "grad_norm": 0.15164400831592006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59896242696, "tokens_per_second": 134923.54183583468, "data_wait_s": 1.2925000191899016e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 409300, "loss": 0.543301522731781, "lr": 0.0003, "grad_norm": 0.17451392721998638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59910878381, "tokens_per_second": 152252.80894552683, "data_wait_s": 7.086000550771132e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f3627755a46d52ee1172c541b17b2a4f0ccf09a32b7c9c1ce5ddf2816bb8270
-size 258009
+oid sha256:12da513ad9e9b74d0abe3ed5fc90d8ebfe36f17812308fc89248036d0b701773
+size 260127