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": 77596.42560977124,
-      "required_pages_per_second_upper_estimate": 0.005368300772197233
+      "required_token_positions_per_second": 76676.38059169205,
+      "required_pages_per_second_upper_estimate": 0.005304649923047956
     },
     {
       "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": 9699.553201221404,
-      "required_pages_per_second_upper_estimate": 0.004061000375647866
+      "required_token_positions_per_second": 9584.547573961507,
+      "required_pages_per_second_upper_estimate": 0.004012849921104738
     },
     {
       "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": 19399.10640244281,
-      "required_pages_per_second_upper_estimate": 0.004040322274790145
+      "required_token_positions_per_second": 19169.095147923013,
+      "required_pages_per_second_upper_estimate": 0.003992416996278361
     },
     {
       "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": 29098.65960366421,
-      "required_pages_per_second_upper_estimate": 0.0017444202121590978
+      "required_token_positions_per_second": 28753.642721884517,
+      "required_pages_per_second_upper_estimate": 0.0017237369769066801
     },
     {
       "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": 29098.65960366421,
-      "required_pages_per_second_upper_estimate": 0.002262292792707852
+      "required_token_positions_per_second": 28753.642721884517,
+      "required_pages_per_second_upper_estimate": 0.0022354692477183622
     },
     {
       "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": 29098.65960366421,
-      "required_pages_per_second_upper_estimate": 0.3732463616893602
+      "required_token_positions_per_second": 28753.642721884517,
+      "required_pages_per_second_upper_estimate": 0.36882085558015565
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10614890920301434,
-    "producer_p95_seconds": 0.19950969700039423,
-    "gpu_update_seconds": 0.9501468581913121,
-    "producer_batches_per_second": 9.420727989653262,
-    "gpu_batches_per_second": 1.0524688803408642,
-    "producer_headroom": 8.951075101244003,
+    "producer_batch_seconds": 0.08135458364054671,
+    "producer_p95_seconds": 0.21170627999890712,
+    "gpu_update_seconds": 0.9615477338791927,
+    "producer_batches_per_second": 12.291870417754865,
+    "gpu_batches_per_second": 1.0399899711329759,
+    "producer_headroom": 11.819220145328876,
     "sustainable": true,
-    "buffer_seconds": 60.80939892424397,
+    "buffer_seconds": 60.57750723438914,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325764.2400640555,
-    "upload_bytes_per_second": 5375765.8584790025,
+    "generation_bytes_per_second": 319541.54772992415,
+    "upload_bytes_per_second": 5344648.847637103,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.90142251435516,
+    "estimated_full_shards_per_hour": 68.56617759631436,
     "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": 1.266042872000071,
-      "bytes": 11864568,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.21107310699972,
       "bytes": 11765224,
@@ -562,6 +556,12 @@
       "bytes": 1089032,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.843757688999176,
+      "bytes": 12452068,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 570800, "loss": 0.5405009984970093, "lr": 0.0003, "grad_norm": 0.19127721466924744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83548371607, "tokens_per_second": 152170.3913998359, "data_wait_s": 1.4620000001741573e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 570900, "loss": 0.6083635091781616, "lr": 0.0003, "grad_norm": 0.17543180640253944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83562877913, "tokens_per_second": 151782.04519515485, "data_wait_s": 8.551000064471737e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 571000, "loss": 0.48727861046791077, "lr": 0.0003, "grad_norm": 0.18006728943270417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83577426920, "tokens_per_second": 148787.13178709097, "data_wait_s": 1.1567000910872594e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 571100, "loss": 0.5819798111915588, "lr": 0.0003, "grad_norm": 0.167887702044046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83592114214, "tokens_per_second": 153179.62478965265, "data_wait_s": 7.991000529727899e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7e479a1c2bbff739721b71d79b4a772a9924d675d04aea7d2ea8006560598e4
-size 296916
+oid sha256:31e9beceab156f90c791acbf440a4d245a34ef5dc6ff1db5bb8b7cdfda4d3f12
+size 299034