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": 76604.7696820516,
-      "required_pages_per_second_upper_estimate": 0.005299695714158826
+      "required_token_positions_per_second": 77571.82091798644,
+      "required_pages_per_second_upper_estimate": 0.00536659856252881
     },
     {
       "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": 9575.59621025645,
-      "required_pages_per_second_upper_estimate": 0.004009102172047161
+      "required_token_positions_per_second": 9696.477614748304,
+      "required_pages_per_second_upper_estimate": 0.004059712691817128
     },
     {
       "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": 19151.1924205129,
-      "required_pages_per_second_upper_estimate": 0.003988688330285505
+      "required_token_positions_per_second": 19392.95522949661,
+      "required_pages_per_second_upper_estimate": 0.0040390411476827915
     },
     {
       "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": 28726.788630769348,
-      "required_pages_per_second_upper_estimate": 0.0017221271151481488
+      "required_token_positions_per_second": 29089.432844244915,
+      "required_pages_per_second_upper_estimate": 0.0017438670820203567
     },
     {
       "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": 28726.788630769348,
-      "required_pages_per_second_upper_estimate": 0.0022333814602528215
+      "required_token_positions_per_second": 29089.432844244915,
+      "required_pages_per_second_upper_estimate": 0.0022615754527472273
     },
     {
       "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": 28726.788630769348,
-      "required_pages_per_second_upper_estimate": 0.36847640013300687
+      "required_token_positions_per_second": 29089.432844244915,
+      "required_pages_per_second_upper_estimate": 0.37312801072645185
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3287598986717626,
-    "producer_p95_seconds": 0.3087185780022992,
-    "gpu_update_seconds": 0.9624465983777297,
-    "producer_batches_per_second": 3.041733508375396,
-    "gpu_batches_per_second": 1.0390186860087294,
-    "producer_headroom": 2.927506068307457,
+    "producer_batch_seconds": 0.10463489057804054,
+    "producer_p95_seconds": 0.3078223850025097,
+    "gpu_update_seconds": 0.9504482314260696,
+    "producer_batches_per_second": 9.557041580257241,
+    "gpu_batches_per_second": 1.0521351578502935,
+    "producer_headroom": 9.083473267620903,
     "sustainable": true,
-    "buffer_seconds": 60.63413569779697,
+    "buffer_seconds": 60.828686811268454,
     "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": 318693.2138233053,
-    "upload_bytes_per_second": 5596308.9169442905,
+    "generation_bytes_per_second": 317560.74948794243,
+    "upload_bytes_per_second": 5676414.786426168,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.3841448881566,
+    "estimated_full_shards_per_hour": 68.14114440420823,
     "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": 0.84063605999836,
-      "bytes": 1225280,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.2510575380001683,
       "bytes": 12042377,
@@ -562,6 +556,12 @@
       "bytes": 13078461,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.3595879119966412,
+      "bytes": 10348219,
+      "pages": 44,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -352,3 +352,4 @@
 {"step": 520200, "loss": 0.592900276184082, "lr": 0.0003, "grad_norm": 0.1695769086647062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76139940545, "tokens_per_second": 143883.7605864952, "data_wait_s": 2.0942999981343746e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 520300, "loss": 0.5820107460021973, "lr": 0.0003, "grad_norm": 0.15467962670824772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76154388078, "tokens_per_second": 150642.8708769923, "data_wait_s": 1.507999695604667e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 520400, "loss": 0.6169607043266296, "lr": 0.0003, "grad_norm": 0.1641458184039181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76168946742, "tokens_per_second": 152582.11370443949, "data_wait_s": 1.3524993846658617e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 520500, "loss": 0.5385158658027649, "lr": 0.0003, "grad_norm": 0.19011648300441572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76183513598, "tokens_per_second": 152017.2329519164, "data_wait_s": 1.3902004866395146e-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:e576f66e8f69bd088b3b2fea0a1e87323e4237485584b9a81dbe7dea61d1e54b
-size 755418
+oid sha256:0320d1c49112dd86a79378b0bd9601f5369e3c8021240fe72b3c7242ab02ab2c
+size 757536