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": 76500.85398979121,
-      "required_pages_per_second_upper_estimate": 0.005292506585450627
+      "required_token_positions_per_second": 75466.9572492238,
+      "required_pages_per_second_upper_estimate": 0.005220979209967221
     },
     {
       "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": 9562.606748723902,
-      "required_pages_per_second_upper_estimate": 0.004003663755754285
+      "required_token_positions_per_second": 9433.369656152974,
+      "required_pages_per_second_upper_estimate": 0.0039495548838722186
     },
     {
       "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": 19125.213497447803,
-      "required_pages_per_second_upper_estimate": 0.003983277605721319
+      "required_token_positions_per_second": 18866.73931230595,
+      "required_pages_per_second_upper_estimate": 0.0039294442493789175
     },
     {
       "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": 28687.820246171705,
-      "required_pages_per_second_upper_estimate": 0.001719791019992798
+      "required_token_positions_per_second": 28300.108968458924,
+      "required_pages_per_second_upper_estimate": 0.00169654832089476
     },
     {
       "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": 28687.820246171705,
-      "required_pages_per_second_upper_estimate": 0.002230351839754164
+      "required_token_positions_per_second": 28300.108968458924,
+      "required_pages_per_second_upper_estimate": 0.002200208993273678
     },
     {
       "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": 28687.820246171705,
-      "required_pages_per_second_upper_estimate": 0.3679765555363798
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28300.108968458924,
+      "required_pages_per_second_upper_estimate": 0.3630034115578164
     }
   ],
   "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": 0.1376842744452631,
-    "producer_p95_seconds": 0.28528834600001574,
-    "gpu_update_seconds": 0.9637539472414094,
-    "producer_batches_per_second": 7.262993570101237,
-    "gpu_batches_per_second": 1.0376092392278538,
-    "producer_headroom": 6.999738721974044,
+    "producer_batch_seconds": 0.19609747470317984,
+    "producer_p95_seconds": 0.2803316610006732,
+    "gpu_update_seconds": 0.9769573689915572,
+    "producer_batches_per_second": 5.099504731071299,
+    "gpu_batches_per_second": 1.0235861172040988,
+    "producer_headroom": 4.981998725227415,
     "sustainable": true,
-    "buffer_seconds": 60.71649867620879,
+    "buffer_seconds": 60.571356877476546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 161,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 628376.1801194692,
-    "upload_bytes_per_second": 5182538.207553222,
+    "generation_bytes_per_second": 602470.9128202068,
+    "upload_bytes_per_second": 5199253.586598584,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.4174502024081,
+    "estimated_full_shards_per_hour": 64.63811654307678,
     "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": 0.9464336839992029,
-      "bytes": 2380354,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.6319502810001723,
       "bytes": 16693932,
@@ -562,6 +556,12 @@
       "bytes": 7938212,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.2239695549978933,
+      "bytes": 5214335,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -322,3 +322,4 @@
 {"step": 752200, "loss": 0.5571590662002563, "lr": 5e-05, "grad_norm": 0.2566676285867222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110100594628, "tokens_per_second": 151454.34595932436, "data_wait_s": 1.3125998520990834e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 752300, "loss": 0.5784069299697876, "lr": 5e-05, "grad_norm": 0.2570707362933527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110115272638, "tokens_per_second": 151708.78269091534, "data_wait_s": 1.6049001715146005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 752400, "loss": 0.5917224287986755, "lr": 5e-05, "grad_norm": 0.29516761130177943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110130060104, "tokens_per_second": 152173.15408398607, "data_wait_s": 8.39999847812578e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 752500, "loss": 0.5992336273193359, "lr": 5e-05, "grad_norm": 0.2487611479791861, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110144784141, "tokens_per_second": 152208.08170043066, "data_wait_s": 1.4673001714982092e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99f392663c086e81a29c218b75992e1d9f8ac6212ee19e77a2c1ad4fbd52ba16
-size 691095
+oid sha256:fd717acc69fb88a937a09fed2e0b320c882a99dde2e7991c45531548fb15c1a0
+size 693213