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": 77064.68612224968,
-      "required_pages_per_second_upper_estimate": 0.005331513800644895
+      "required_token_positions_per_second": 76043.67484817725,
+      "required_pages_per_second_upper_estimate": 0.005260877871632015
     },
     {
       "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": 9633.08576528121,
-      "required_pages_per_second_upper_estimate": 0.004033171848217603
+      "required_token_positions_per_second": 9505.459356022156,
+      "required_pages_per_second_upper_estimate": 0.003979737335803327
     },
     {
       "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": 19266.17153056242,
-      "required_pages_per_second_upper_estimate": 0.004012635446705782
+      "required_token_positions_per_second": 19010.918712044313,
+      "required_pages_per_second_upper_estimate": 0.0039594730160779565
     },
     {
       "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": 28899.25729584363,
-      "required_pages_per_second_upper_estimate": 0.0017324663482749484
+      "required_token_positions_per_second": 28516.378068066468,
+      "required_pages_per_second_upper_estimate": 0.0017095133231924376
     },
     {
       "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": 28899.25729584363,
-      "required_pages_per_second_upper_estimate": 0.0022467901403528615
+      "required_token_positions_per_second": 28516.378068066468,
+      "required_pages_per_second_upper_estimate": 0.0022170229645009283
     },
     {
       "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": 28899.25729584363,
-      "required_pages_per_second_upper_estimate": 0.370688642986155
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28516.378068066468,
+      "required_pages_per_second_upper_estimate": 0.3657774793559147
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11441918854694677,
-    "producer_p95_seconds": 0.33635060799861094,
-    "gpu_update_seconds": 0.9567027870981449,
-    "producer_batches_per_second": 8.739792797863576,
-    "gpu_batches_per_second": 1.0452567019619368,
-    "producer_headroom": 8.361384128376377,
+    "producer_batch_seconds": 0.14586033255449138,
+    "producer_p95_seconds": 0.33515112299937755,
+    "gpu_update_seconds": 0.9695480938710479,
+    "producer_batches_per_second": 6.855873577735153,
+    "gpu_batches_per_second": 1.0314083502628208,
+    "producer_headroom": 6.647099159113999,
     "sustainable": true,
-    "buffer_seconds": 60.27227558718313,
+    "buffer_seconds": 60.11198182000497,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451913.87356369593,
-    "upload_bytes_per_second": 4548321.311453781,
+    "generation_bytes_per_second": 454302.00789400976,
+    "upload_bytes_per_second": 4489788.073212042,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.9701972502056,
+    "estimated_full_shards_per_hour": 97.48263528456897,
     "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": 1.1368510719985352,
-      "bytes": 7197379,
-      "pages": 42,
-      "shards": 2
-    },
     {
       "seconds": 1.1000944999977946,
       "bytes": 5635539,
@@ -562,6 +556,12 @@
       "bytes": 10042142,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 1.317283235002833,
+      "bytes": 2476726,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -427,3 +427,4 @@
 {"step": 802700, "loss": 0.5637000203132629, "lr": 3e-05, "grad_norm": 0.2688420309560021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117492060387, "tokens_per_second": 152672.75408613237, "data_wait_s": 8.587005140725523e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 802800, "loss": 0.545868456363678, "lr": 3e-05, "grad_norm": 0.2683358329979781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117506774323, "tokens_per_second": 152528.6755177442, "data_wait_s": 1.4285993529483676e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802900, "loss": 0.5391790270805359, "lr": 3e-05, "grad_norm": 0.2842515022638538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117521292308, "tokens_per_second": 150410.5781571794, "data_wait_s": 2.0007006241939962e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 803000, "loss": 0.5340162515640259, "lr": 3e-05, "grad_norm": 0.274242308446818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117535845279, "tokens_per_second": 151410.89411223173, "data_wait_s": 1.251199864782393e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:695b75ac4fb9c94d0d22bfbe1e9916edfa21da814d8127ab87b247b7217a3646
-size 915051
+oid sha256:95273356fc9ce8d904868f7140dfe4fa5294abb5f3bd9e661aee333bf8477219
+size 917169