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": 77462.6587097821,
-      "required_pages_per_second_upper_estimate": 0.005359046467674014
+      "required_token_positions_per_second": 77331.14923659828,
+      "required_pages_per_second_upper_estimate": 0.00534994833717514
     },
     {
       "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": 9682.832338722763,
-      "required_pages_per_second_upper_estimate": 0.0040539996996909996
+      "required_token_positions_per_second": 9666.393654574786,
+      "required_pages_per_second_upper_estimate": 0.004047117165917002
     },
     {
       "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": 19365.664677445526,
-      "required_pages_per_second_upper_estimate": 0.004033357245390806
+      "required_token_positions_per_second": 19332.78730914957,
+      "required_pages_per_second_upper_estimate": 0.004026509756609266
     },
     {
       "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": 29048.49701616829,
-      "required_pages_per_second_upper_estimate": 0.0017414130416325533
+      "required_token_positions_per_second": 28999.18096372436,
+      "required_pages_per_second_upper_estimate": 0.0017384566195897918
     },
     {
       "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": 29048.49701616829,
-      "required_pages_per_second_upper_estimate": 0.0022583928721719485
+      "required_token_positions_per_second": 28999.18096372436,
+      "required_pages_per_second_upper_estimate": 0.002254558766012823
     },
     {
       "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": 29048.49701616829,
-      "required_pages_per_second_upper_estimate": 0.37260292987735266
+      "required_token_positions_per_second": 28999.18096372436,
+      "required_pages_per_second_upper_estimate": 0.37197035650805355
     }
   ],
   "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": 0.10073907532802195,
-    "producer_p95_seconds": 0.2153555469994899,
-    "gpu_update_seconds": 0.9517876255219408,
-    "producer_batches_per_second": 9.92663469208791,
-    "gpu_batches_per_second": 1.050654550642661,
-    "producer_headroom": 9.448048063006075,
+    "producer_batch_seconds": 0.4967029945312049,
+    "producer_p95_seconds": 0.21351718799996888,
+    "gpu_update_seconds": 0.9534062370446056,
+    "producer_batches_per_second": 2.013275561070079,
+    "gpu_batches_per_second": 1.0488708392550767,
+    "producer_headroom": 1.919469476813691,
     "sustainable": true,
-    "buffer_seconds": 60.914408033404214,
+    "buffer_seconds": 60.064592933810154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285806.39791102736,
-    "upload_bytes_per_second": 5519044.6249654535,
+    "generation_bytes_per_second": 267250.36055005883,
+    "upload_bytes_per_second": 5645646.442275783,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.327399759274634,
+    "estimated_full_shards_per_hour": 57.345706104052766,
     "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": 1.9445103569996718,
-      "bytes": 2479005,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.5074067269997613,
       "bytes": 15537281,
@@ -562,6 +556,12 @@
       "bytes": 7688939,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.4415530749975005,
+      "bytes": 9558241,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -193,3 +193,4 @@
 {"step": 544300, "loss": 0.5637333989143372, "lr": 0.0003, "grad_norm": 0.15272475208395467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79668280993, "tokens_per_second": 155063.12250591084, "data_wait_s": 1.5093999536475167e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 544400, "loss": 0.5742185115814209, "lr": 0.0003, "grad_norm": 0.15374521942284794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79682963821, "tokens_per_second": 153595.49709699335, "data_wait_s": 9.047998901223764e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 544500, "loss": 0.5352970361709595, "lr": 0.0003, "grad_norm": 0.1683381556944048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79697516793, "tokens_per_second": 148879.42333393815, "data_wait_s": 1.5584002539981157e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 544600, "loss": 0.5824766159057617, "lr": 0.0003, "grad_norm": 0.1549186194981331, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79712074304, "tokens_per_second": 152070.29173009028, "data_wait_s": 1.1586998880375177e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa42ebb0aedfa1af0dd2e43ee9c9d337c7367d37ed0f4dce27af22c29efa5df6
-size 415524
+oid sha256:2756f2878f5e0801707285dc6cc8a55a1c0c792922396fff69de8ad56179ae79
+size 417642